Editorial note: This publication and Quirky Cotton Collective share an owner.

The garment preview shows four characters and a period. Lowercase. Centered. Very small on a dark shirt. No animal. No slogan about coffee, spreadsheets, or hating people. Just meh. — and the period is doing more work than the word.

I want to argue that this is the most honest piece of typography in the entire introvert-merch category, and that its honesty is exactly why it fails as a wearable message. That failure is the interesting part.

What is actually printed

Observed on the native garment preview: the word meh in lowercase, followed by a period, set very small and centered on a dark shirt. No illustration. No secondary text. No animal, no pride variant, no flag. The retail title adds Funny Sarcastic Introvert Minimalist Saying, but those are category words, not printed words. The shirt itself says one thing.

That distinction matters. A lot of shirts in this space carry a paragraph. This one carries a syllable and a punctuation mark. The design decision is subtraction, and subtraction is harder to evaluate than addition because there is less to argue with.

The period is the whole design

Strip the period and you have a mood. meh floating on a chest is ambient. It could be a brand, a band, a typo. The period converts it into a statement — a complete sentence, grammatically closed, delivered flat. In typography, a period is a full stop. On a shirt, a full stop is a refusal to continue the conversation.

That is a real observation about the artwork, not a claim about the wearer. I am reading the mark, not the person.

Scale reinforces it. Very small, centered lettering reads as understatement rather than announcement. A large MEH across the chest would be a joke shouted. A small meh. is a joke muttered. The design chooses the mutter.

Why it works as an object and fails as a message

Here is my argument. The shirt is a good object because it is legible at the distance where it matters — close enough to read, far enough to be intriguing — and it commits to one idea without decoration. It is a clean piece of minimal typography.

It is a poor message because meh. has no addressee. Sarcasm needs a target. Introvert humor usually needs a situation — a party, a meeting, a group chat. This shirt names neither. It expresses a stance toward everything in general, which in practice means nothing in particular. A stranger who reads it learns that you are unimpressed, but not by what, and not in a way they can respond to. The period closes the door before anyone knocks.

That is not a flaw in the printing. It is a flaw in the premise, and the premise is the product. If you want a shirt that starts conversations, this is the wrong shirt. If you want a shirt that ends them, the design is unusually well calibrated.

The introvert label is doing unpaid labor

The listing calls this an introvert saying. The artwork does not. meh. is not introversion; it is indifference, which is a different thing and arguably a more universal one. Introversion is about where you get your energy. Indifference is about whether you care. The shirt prints the second and the listing sells the first.

I think that mismatch is why the design survives at all. If the shirt literally said I am an introvert, it would be a label, and labels age badly. meh. is a mood, and moods are reusable. The retail framing borrows the audience; the artwork keeps its options open.

What the demand data does and does not say

Google Keyword Planner data for the US, September 2025 through August 2026, puts the exact keyword meh shirt in the 100–1K monthly range, with 0% change over three months and 0% over the year, and high ad competition. Related queries include meh introvert shirt and meh sarcastic shirt, but those are retained as variants, not summed into a total. I am not going to pretend a flat year-over-year number is a trend story. It is a floor, not a forecast.

What the data supports is narrow: people search this phrase at a low but nonzero rate, consistently, without seasonal spikes. The shirt is not a fad and not a wave. It is a small steady object, which is exactly what the design claims to be.

If you want the specific garment, it is listed here: Meh Funny Sarcastic Introvert Minimalist Saying.

The maintenance burden of understatement

Every design choice creates a cost. The cost of a very small, centered, lowercase meh. is legibility at distance. From across a room, the shirt reads as a dark rectangle with a smudge. The joke only lands inside a narrow band of proximity — close enough to parse four letters, far enough that the wearer is not explaining them. Outside that band, the shirt is just fabric.

That is the tradeoff. You get a design that never overexplains and never gets old, and you pay for it in reach. A louder shirt would be read by more people and regretted by the wearer sooner. This one is read by fewer people and regretted less. I do not think there is a version that avoids both costs. The period guarantees it.

So the honest summary is this: the artwork is a small, well-executed refusal, and refusals do not scale. If that is what you want on a shirt, the execution is clean. If you wanted a conversation starter, you bought a conversation ender, and the design told you so in four characters and a dot.

Editorial note: This publication and Quirky Cotton Collective share an owner.

A white rounded notification banner with a bell reads There is motion at your front door. Below it, a circular, green-tinted doorbell-camera view shows a wide-eyed orange tabby very close to the lens, porch siding and lights behind it. That is the whole design. No caption, no punchline beyond the image itself.

The joke works because the notification is technically correct and operationally useless. Motion was detected. The system did its job. The human receiving the alert now has to decide whether the motion matters. It does not. The cat is not a package, a visitor, or a threat. The alert fired anyway.

Anyone who has owned a pager, a Slack alert channel, or a doorbell camera knows this exact failure mode. The sensor is not wrong. The threshold is wrong. The notification is not broken. The definition of “motion” is too broad to be actionable.

The shirt is available at the Amazon listing under the title “There Is Motion At Your Front Door Funny Cat Doorbell Meme.” The design is a direct visual quotation of a doorbell-camera push notification, with the cat standing in for the usual delivery driver or raccoon. The green tint is the giveaway: that is the night-vision or low-light cast of a cheap camera sensor, not a stylistic choice. The circular crop is the fisheye lens. The banner is the operating system notification. Every element is a faithful reproduction of the interface, which is why the cat reads as an intruder rather than a pet.

The shirt is not making a technical argument. It is making a cultural one: we have all built systems that page us for the equivalent of a cat walking past a camera. The cat is not the problem. The problem is that we configured the camera to treat every pixel change as an event worth waking someone up for.

There is a version of this joke that would be smug about it. The shirt is not that. The cat looks startled, not mischievous. The notification is neutral. The composition puts the viewer in the position of the person checking their phone at 2 a.m., seeing “motion at your front door,” and feeling the small spike of adrenaline before the image loads. That spike is the real subject. The cat is just the payload.

If you work in operations, you have felt that spike. It is the same spike you feel when a PagerDuty alert fires for a disk usage threshold that has been at 81% for six months. The alert is not lying. The disk is above the threshold. But the threshold was set by someone who did not think about what happens when the disk sits at 81% forever. The alert becomes noise. The cat becomes the doorbell.

The design does not offer a fix. It does not need to. It is a shirt. But it does capture something specific about the gap between detection and meaning. A doorbell camera detects motion. A human decides whether the motion matters. The shirt puts the cat in the frame to remind you that the detection layer and the meaning layer are not the same thing. The cat is motion. The cat is not a reason to open the door.

I do not know whether the shirt is comfortable. I have not worn it. I am not going to pretend that the fabric weight or the print durability is the point. The point is the image. The image is a notification that has been rendered absurd by its own success. The system worked. The system told you there was motion. The system did not tell you the motion was a cat. That is the gap.

If you are looking for a shirt that says something about alert fatigue without saying “alert fatigue,” this is it. The phrase “there is motion at your front door shirt” describes the object, but the object is really about the moment after the notification arrives. The moment when you look at your phone and realize the system has done exactly what you asked it to do, and that is the problem.

The cat is not sorry. The cat is not even aware of the camera. The cat is just being a cat. The notification is just being a notification. The humor is in the collision. That collision is familiar to anyone who has ever tuned a threshold, written an alert rule, or explained to a stakeholder why the system paged them for a non-event. The shirt does not solve that problem. It just names it, in the form of a wide-eyed orange tabby who has no idea they have become a false positive.

A field is deleted from a Protobuf message. The number is not reserved. Months later, a new field takes that number. The schema compiles. The registry’s compatibility check returns a verdict. Old bytes still sit in Kafka topics, S3 archives, and downstream caches. When those bytes are decoded under the new definition, the wire format has no way to tell you the number means something different now.

This is not a hypothetical about a specific incident. It is the documented failure shape of field-number reuse, and it is worth understanding precisely because the tooling that looks like it should catch it does not.

What the wire format actually carries

A Protobuf message on the wire is a series of key-value pairs. The key is the field number plus a wire type. The name and declared type of the field are not on the wire at all.

The Protobuf encoding guide states it directly: “The binary version of a message just uses the field’s number as the key – the name and declared type for each field can only be determined on the decoding end by referencing the message type’s definition (i.e. the .proto file).”

That design is what makes Protobuf lean and forward-compatible. It is also what makes reuse undetectable at the byte level. If field 14 was once an int32 and is now a string, the decoder does not know. It reads the tag, sees field 14, and applies whatever the current .proto says field 14 is.

The language guide is explicit about the consequence: “Reusing a field number makes decoding wire-format messages ambiguous. The Protobuf wire format is lean and doesn’t provide a way to detect fields encoded using one definition and decoded using another.”

There is no checksum, no type tag, no version marker. The number is the identity.

The rule the docs give, and the procedure they prescribe

The proto3 language guide is unambiguous: “Field numbers should never be reused. Never take a field number out of the reserved list for reuse with a new field definition.”

It also explains why the number cannot simply be changed: “This number cannot be changed once your message type is in use because it identifies the field in the message wire format.”

When a field is deleted, the documented procedure is to reserve the number. From the same guide: “However, you must reserve the deleted field number. If you do not reserve the field number, it is possible for a developer to reuse that number in the future.”

The mechanism is a compiler check. “The protoc compiler will generate error messages if any future developers try to use these reserved field numbers.”

Two syntax details matter in practice. Reserved ranges are inclusive: reserved 9 to 11; covers 9, 10, and 11. And field names and field numbers cannot be mixed in the same reserved statement — you need separate lines.

message OrderEvent {
  reserved 14;
  reserved "legacy_customer_id";
  // ...
}

The best-practices page reinforces the same point with a blunt rationale: “Never re-use a tag number. It messes up deserialization. Even if you think no one is using the field, don’t re-use a tag number. If the change was live ever, there could be serialized versions of your proto in a log somewhere. Or there could be old code in another server that will break.”

And on the deletion side: “When you delete a field that’s no longer used, reserve its tag number so that no one accidentally re-uses it in the future. Just reserved 2, 3; is enough.”

What the compatibility check actually compares

Confluent Schema Registry enforces compatibility by comparing new schema versions against previous versions using configurable compatibility types. The default is BACKWARD. The documented compatibility types are:

• Backward — consumers using the new schema can read data written with the old schema (add optional fields, remove fields).
• Forward — consumers using the old schema can read data written with the new schema (remove optional fields, add fields).
• Full — both backward and forward compatible (add/remove optional fields only).
• Transitive — compatibility checked against all previous versions, not just the last one.

This is a schema-diff against a policy. It answers the question: does this new schema version satisfy the configured compatibility rule relative to prior versions? It is not a scan of historical bytes for a number that was previously assigned to a different field. The registry documentation describes compatibility policy, not reuse detection.

The allowed changes also depend on how fields were originally defined. The Schema Registry docs note: “The allowed changes are dependent on how fields are originally defined. For example, as described under Backward compatibility, the ability to delete a field and keep the schema compatible requires that the field was either specified as optional or provided a default value in the original version.”

So a deletion can pass the check. A subsequent addition of a new field with the same number can also pass the check, because the check is comparing schema versions against a policy, not auditing the number’s history. The compatibility verdict and the wire-format safety question are different questions.

Confluent’s own guidance for Protobuf is to use BACKWARD_TRANSITIVE, “as adding new message types is not forward compatible.” That is a stronger policy, but it still does not turn the check into a reuse detector. It compares more versions; it does not know that a number was once something else.

The failure shape, illustrated

Consider a message where field 14 was an int32 representing a legacy customer ID. The field is removed. No reserved 14; line is added. The PR merges.

Months later, a new field is added: string region = 14;. The schema compiles. The compatibility check runs against the prior version and returns a verdict based on the configured policy. Nothing in that check knows field 14 used to be an int32.

Old records still carrying the original field 14 bytes now decode under the new definition. Depending on wire-type compatibility, the result is either a parse/merge error or a silently wrong value. The language guide lists the possible outcomes of encoding with one definition and decoding with another: “Developer time lost to debugging. A parse/merge error (best case scenario). Leaked PII/SPII. Data corruption.”

The parse error is the good outcome. The silent wrong value is the one that reaches a dashboard or a downstream join.

This is a hypothetical illustration of the documented failure shape, not a report of a specific observed incident.

The fix is a line in the deletion change

The cheapest control point is the deletion itself. If reserved <number>; is not part of the same commit that removes the field, the number is available for reuse the moment the change merges. No later tooling reliably reconstructs the intent.

Treat the reserved line as a required artifact of a field deletion, reviewed the same way as the field removal. A practical review heuristic:

• Any diff that removes a field definition should also add a reserved entry for that number.
• Any diff that adds a field should be checked against the reserved list before merge.

This is a process argument, not a claim about any specific tool’s behavior. The compiler enforces the reserved list once it exists. The gap is between deletion and reservation.

The cost of the fix is minutes at deletion time. The cost of skipping it is debugging time, parse/merge errors, leaked PII/SPII, or data corruption discovered later — potentially after the old bytes have been replayed, backfilled, or joined into a downstream table.

Why this keeps happening

The language guide names the common causes of field number reuse: “renumbering fields (sometimes done to achieve a more aesthetically pleasing number order for fields). Renumbering effectively deletes and re-adds all the fields involved in the renumbering, resulting in incompatible wire-format changes. deleting a field and not reserving the number to prevent future reuse.”

The first cause is cosmetic. The second is a missing line in a PR. Neither looks dangerous at the time.

The maintenance tax lands on the on-call engineer who has to trace a corrupted value back through a schema history that was never recorded in the .proto file. The reserved clause is the record. It is the only place the intent survives.

FAQ

Does the Confluent Schema Registry compatibility check detect a reused field number?

The retrieved Schema Registry documentation describes compatibility as a comparison of new schema versions against previous versions using configurable compatibility types. It does not describe a mechanism for detecting that a field number was previously assigned to a different field. The check evaluates allowed schema changes per format and policy; reuse detection is not part of that description.

What happens if I reuse a number that was never reserved?

The schema compiles. The wire format has no way to detect that the number means something different now. Old bytes decode under the new definition, which can produce a parse/merge error or a silently wrong value, depending on wire-type compatibility. The language guide lists developer time lost to debugging, parse/merge errors, leaked PII/SPII, and data corruption as possible outcomes.

Can I just change the field number instead of reserving it?

The language guide states that the number cannot be changed once the message type is in use because it identifies the field in the wire format. “Changing” a field number is equivalent to deleting that field and creating a new field with the same type but a new number. The documented procedure for deletion is to reserve the old number.

Do reserved names affect runtime JSON parsing?

The language guide states that reserved names affect only the protoc compiler behavior and not runtime behavior, with one exception: TextProto implementations may discard unknown fields with reserved names at parse time (only the C++ and Go implementations do so today). Runtime JSON parsing is not affected by reserved names.

What is the recommended compatibility mode for Protobuf?

The Confluent Schema Registry documentation notes that the best practice for Protobuf is to use BACKWARD_TRANSITIVE, because adding new message types is not forward compatible. The default compatibility type is BACKWARD.

Sources

• Protobuf Language Guide (proto3), “Assigning Field Numbers,” “Consequences of Reusing Field Numbers,” “Deleting Fields,” “Reserved Field Numbers,” “Reserved Field Names”: https://protobuf.dev/programming-guides/proto3/#reserved
• Protobuf Encoding Guide, “Message Structure”: https://protobuf.dev/programming-guides/encoding/#structure
• Protobuf Best Practices, “Don’t Re-use a Tag Number,” “Do Reserve Tag Numbers for Deleted Fields”: https://protobuf.dev/best-practices/dos-donts/
• Confluent Schema Registry, “Schema Evolution and Compatibility”: https://docs.confluent.io/platform/current/schema-registry/fundamentals/schema-evolution.html

Somewhere in your Slack workspace there is a channel called #data-alerts. It fired forty times yesterday. Nobody read it. The one message that mattered — a source table that stopped receiving CDC events at 02:14 and stayed stale for six hours — scrolled past under thirty-nine “source X is 61 minutes stale” notifications that resolved themselves within two minutes.

That channel is not a monitoring failure. It is a design failure. The alert was written to answer the wrong question.

A static freshness threshold answers: is this one file late right now? The operational question is: are we on pace to miss the SLA? Those are different questions, and only one of them deserves a page.

What a static threshold actually measures

Take a dbt source with warn_after: {count: 1, period: hour} and error_after: {count: 2, period: hour}. The check compares the max load timestamp in the source against wall-clock time. If the upstream extractor runs at 02:00 and finishes at 02:03, the check at 02:05 passes. If it finishes at 03:01, the check fails. The alert fires on a single late file.

The Google SRE Workbook makes the cost of this pattern explicit. In its first alerting iteration — alert when the recent error rate equals the SLO over a short window — the authors note you could receive up to 144 alerts per day every day, not act upon any alerts, and still meet the SLO (SRE Workbook, Chapter 5). That is the arithmetic of a short window: excellent detection time, terrible precision. Every late file crosses the line. Almost none of them threaten the SLA.

The obvious fix — widen the window — has its own failure mode. The Workbook’s second iteration uses a 36-hour window to ensure only sustained problems alert. Precision improves. Reset time collapses: in the case of 100% outage, an alert will fire shortly after 2 minutes, and continue to fire for the next 36 hours. You have traded a noisy channel for a channel that lies to you for a day and a half after the incident is over.

The third iteration — add a for: 1h duration — is worse. The Workbook is blunt: because the duration does not scale with the severity of the incident, a 100% outage alerts after one hour, the same detection time as a 0.2% outage. A service that spikes to 100% errors for five minutes every ten minutes never triggers the alert at all, despite consuming 35% of the monthly budget. Duration parameters do not measure severity. They measure persistence.

Reframe freshness as an SLO with a budget

If your freshness SLA is “data is at most one hour old,” you have already defined an SLO. The good event is a load that arrives within the SLA. The bad event is a load that arrives late. The error budget is the fraction of loads you are allowed to deliver late over the measurement window — typically 30 days.

Once you frame it that way, a single late file is not an incident. It is budget spend. The pager should fire when the rate of budget spend threatens to exhaust the budget before the window closes. That rate is the burn rate.

Burn rate 1 means you are consuming budget at exactly the pace that exhausts it at the end of the window. Burn rate 2 exhausts it in half the time. The Workbook’s table for a 99.9% SLO over 30 days: burn rate 1 corresponds to a 0.1% error rate and 30 days to exhaustion; burn rate 10 corresponds to 1% and 3 days; burn rate 1,000 corresponds to 100% and 43 minutes.

For freshness, the mapping is direct. If your SLA is one hour and your measurement window is 30 days, then a source that is late 0.1% of the time is burning at rate 1. A source that is late 1% of the time is burning at rate 10 and will exhaust the budget in three days. A source that stops entirely is burning at rate 1,000 and will exhaust it in 43 minutes.

The rule shape that replaces the threshold

The Workbook recommends multiwindow, multi-burn-rate alerts as the most viable option. The recommended starting numbers: 2% budget consumption in one hour and 5% budget consumption in six hours as reasonable starting numbers for paging, and 10% budget consumption in three days as a good baseline for ticket alerts.

Translated into burn rates for a 30-day window:

• Page: 14.4x burn rate over 1 hour, confirmed by 14.4x over 5 minutes. This is 2% of the budget in one hour.
• Page: 6x burn rate over 6 hours, confirmed by 6x over 30 minutes. This is 5% of the budget in six hours.
• Ticket: 1x burn rate over 3 days. This is 10% of the budget in three days.

The short confirmation window is the key mechanism. The Workbook’s guideline: make the short window 1/12 the duration of the long window. The long window establishes that a significant amount of budget has been spent. The short window establishes that the budget is still being spent. If the short window has recovered, the alert does not fire — which is exactly the class of alert that trains people to mute the channel.

In Prometheus, this maps onto the alerting rule primitives directly. The for clause waits a duration before firing; the keep_firing_for clause keeps the alert firing after the condition was last met, which the documentation describes as useful to prevent situations such as flapping alerts, false resolutions due to lack of data loss, etc. (Prometheus alerting rules). The same page is explicit that Prometheus alerting rules are not a notification solution: another layer is needed to add summarization, notification rate limiting, silencing and alert dependencies. That layer is Alertmanager. If you skip it, you will rebuild rate limiting and silencing by hand, badly.

Wiring it to the actual stack

dbt source freshness and Prometheus burn-rate rules are complementary, not competing. dbt produces the SLI — how stale is the source. Prometheus produces the alerting logic — how fast is the budget burning.

Two dbt behaviors matter for this design. First, dbt build does not include source freshness checks; you either select the “Run source freshness” checkbox in the job, which runs it as the first step and won’t break subsequent steps if it fails, or you add dbt source freshness as a run step, in which case if your source data is out of date — this step will “fail”, and subsequent steps will not run (dbt source freshness docs). The checkbox is the right choice if you want freshness to feed an alerting pipeline without blocking the models. The run step is the right choice if you want stale data to halt the DAG. Pick one deliberately; the default behavior of the run step has surprised more than one on-call engineer at 03:00.

Second, check frequency is not optional. dbt’s own guidance: you should run your source freshness jobs with at least double the frequency of your lowest SLA. If your SLA is one hour, run the check every 30 minutes. A daily freshness check against a one-hour SLA measures nothing useful — it tells you whether the source was fresh at the moment you happened to look.

The recording rule that turns freshness into a ratio is straightforward. Emit a gauge per source: 1 if fresh, 0 if stale. Then:

record: source:freshness_ratio_rate1h
expr: sum(rate(source_fresh[1h])) by (source)
      / sum(rate(source_expected[1h])) by (source)

The alert then compares that ratio against the burn-rate threshold. For a 99.9% freshness SLO, the 14.4x page rule is source:freshness_ratio_rate1h < (1 - 14.4 * 0.001) combined with the 5-minute confirmation. The exact numbers depend on your SLA; the shape does not.

Where the freshness SLI lies to you

A freshness check that only reads the max timestamp in the target table can miss a permanently skipped file. Snowpipe is the clearest example. Its file-loading metadata is maintained for 14 days. Files that failed to load — because of invalid content or stage access failures — are still registered in the pipe’s metadata, and the registered file names are ignored by subsequent pipe activity, including ALTER PIPE … REFRESH (Snowpipe troubleshooting). The target table’s max timestamp can look perfectly fresh while a specific partition is permanently missing rows.

The same page documents the inverse trap: files modified and staged again after 14 days are loaded again, potentially duplicating records. A freshness SLI that only measures recency will not catch either case. Pair it with a load-history check — COPY_HISTORY for status, SYSTEM$PIPE_STATUS for lastReceivedMessageTimestamp versus lastForwardedMessageTimestamp. The gap between those two timestamps distinguishes a service configuration problem from a path mismatch between the stage and pipe definitions. A freshness alert alone cannot make that distinction.

What the fix costs

Define the SLO per source: one hour of work per source if the SLA is already documented, half a day if it is not. Write the recording rule and the three alert rules: two to four hours for the first source, thirty minutes for each subsequent one once the pattern is templated. Wire Alertmanager routing and suppression: half a day, once.

The ongoing maintenance tax is real and should be priced honestly. You now have more windows, more thresholds, and more numbers to reason about. The Workbook names this directly as a disadvantage of multi-burn-rate alerting. The three-day ticket window also produces a longer reset time than a short-window alert would. And you need alert suppression, because a 10% budget spend in five minutes also means 5% was spent in six hours and 2% in one hour — three conditions true, three notifications, unless the monitoring system prevents it.

The trade is fewer pages and a ticket queue that catches slow burns. For an on-call engineer who was about to mute the channel, that is usually the right trade. It is not free, and it is not a platform migration — it is a few hours per source plus a routing layer you probably already have.

The operational rule

The pager should fire on budget spend, not on a single late file. The ticket queue catches the slow burns that would otherwise exhaust the budget unnoticed. The muted channel is the symptom; the static threshold is the cause.

If you want to test this on one source before committing: pick the noisiest freshness alert you have, count how many times it fired last week, and count how many of those firings corresponded to a real SLA miss. If the ratio is worse than 10:1, the threshold is measuring the wrong thing. Replace it with a burn-rate rule and watch the channel go quiet — not because you muted it, but because it stopped lying to you.

FAQ

Do I need Prometheus to do this? No. The Workbook’s examples use Prometheus syntax, but the page states the approach applies in any alerting framework. What you need is a way to compute a ratio over a window and compare it to a threshold, plus a notification layer that can suppress and route. Most observability platforms have both.

What if my SLA is not a clean number like 99.9%? The burn rate is derived from the SLA, not the other way around. If your SLA is 99% over 30 days, the error budget is 1%, and burn rate 1 corresponds to a 1% error rate. The recommended budget-consumption percentages (2% in 1h, 5% in 6h, 10% in 3d) stay the same; the burn-rate multipliers change.

Should I keep the old static threshold as a backstop? Only if you route it to a ticket queue, not a pager. A static threshold that pages is the problem you are trying to solve. A static threshold that opens a ticket is a cheap safety net for the case where your SLI pipeline itself breaks.

How do I handle sources with no fixed SLA? You cannot burn-rate alert on an undefined budget. Either define the SLA — even a loose one — or route the source to a dashboard and a weekly review. Paging on an undefined SLA is how you get forty alerts a day.

Your reconciliation job reports green. Row counts match on every table. Finance closes the month, and a controller finds a variance that traces back to a column you have been loading for two years.

The failure class is silent decimal coercion. A source column declared numeric with no scale feeds a warehouse column declared NUMBER(18,2). Every value with more than two fractional digits is rounded on write. The row count is unchanged. The money is not.

This is not a pipeline bug in the usual sense. Nothing errored. Nothing was dropped. The type contract between source and warehouse was never tested, because count(*) cannot see it.

What precision and scale actually mean

In Snowflake, precision is the total number of digits allowed and scale is the number of digits allowed to the right of the decimal point. NUMBER defaults to precision 38 and scale 0, i.e. NUMBER(38,0) (Snowflake numeric data types).

In PostgreSQL, the same terms apply. The numeric type is recommended for storing monetary amounts and other quantities where exactness is required (PostgreSQL numeric types).

The asymmetry that matters: an unconstrained numeric column does not coerce input values to any particular scale, whereas a numeric column with a declared scale does coerce input values to that scale. If the scale of a value to be stored is greater than the declared scale of the column, the system rounds the value to the specified number of fractional digits. If the number of digits to the left of the decimal point exceeds the declared precision minus the declared scale, an error is raised.

So a source column with no declared scale feeding a warehouse column with a declared scale is a rounding operation, not a copy. The rounding is deterministic and silent. It does not raise. It does not warn. It does not change the row count.

Why row-count reconciliation is structurally blind

A row-count check verifies cardinality. It answers one question: did the same number of rows arrive? It says nothing about value fidelity.

Consider a payments table. The source column is numeric with no scale. The warehouse column is NUMBER(18,2). A refund of 0.005 rounds to 0.01 in the warehouse and stays 0.005 in the source. Row counts match. The sum differs by a fraction of a cent per row. This is illustrative, not a reported incident.

The same blindness applies to floating-point paths. Snowflake’s FLOAT type uses double-precision (64 bit) IEEE 754 floating-point numbers with precision of approximately 15 digits. Snowflake recommends comparing two floating-point numbers for approximate equality rather than exact equality. PostgreSQL’s real and double precision types are inexact, variable-precision numeric types, and comparing two floating-point values for equality might not always work as expected.

If your reconciliation compares count(*) and nothing else, you are testing the one property that decimal coercion does not change.

The reconciliation query pattern

Compare three things per money column per key range and time window: row count, SUM of the column, and a scale probe. The scale probe is the part most teams skip.

SUM catches aggregate drift. It does not catch offsetting errors. One row rounded up and another rounded down can net to the same total. The scale probe catches the case where rounding is uniform enough that SUM happens to match.

A scale probe asks: what is the maximum number of digits actually present to the right of the decimal point in this column, over this key range? If the source returns 4 and the warehouse returns 2, the warehouse has coerced. The exact expression depends on your engine; the point is to measure the stored scale, not the declared scale.

For per-row fidelity, add a hash or checksum of the money column to the same test. SUM plus scale probe plus per-row hash is one test, not three. Splitting them across separate jobs means a failure in one does not block the others, and the signal gets lost in the noise.

In dbt, this is a singular data test. Data tests are assertions about models and other resources, and a test passes when it returns zero failing rows. Data tests return one row for each failure, and the columns in the test’s SQL select statement are the columns visible when you debug failures, including when you store test failures (dbt data tests).

Write the test so it returns the key range, the source sum, the warehouse sum, the source scale, and the warehouse scale. When it fails, you want the numbers in the row, not a boolean.

The type contract is the artifact under test

The reconciliation query is testing a contract: the source column’s precision and scale must be compatible with the warehouse column’s precision and scale. If the source is unconstrained and the warehouse is declared, the contract is lossy by construction.

Pair the reconciliation query with a schema-diff check that fails when the two sides’ precision and scale declarations diverge. The schema diff catches the drift at deploy time. The reconciliation query catches it at data time. You need both because a schema diff cannot see values that were already rounded before the schema changed.

This is where the decision gets expensive. Snowflake’s DECFLOAT type stores numbers exactly, with up to 38 significant digits of precision, and uses a dynamic base-10 exponent to represent very large or small values. Snowflake lists ledgers, taxes, or compliance as use cases requiring exact numeric values, and notes that use of the DECFLOAT type might cause storage consumption to increase. The NUMBER and FLOAT types might provide better performance than the DECFLOAT type.

So the choice is not free. DECFLOAT buys exactness and costs storage and possibly performance. NUMBER with an explicit scale buys predictability and costs the fractional digits you did not declare. FLOAT buys range and costs exactness. There is no option that is free on all three axes.

CDC and backfill make it worse in a specific way

Debezium’s PostgreSQL connector relies on logical decoding, which does not support DDL changes. The connector is unable to report DDL change events back to consumers (Debezium PostgreSQL connector).

This means a column type change on the source is invisible to the connector. The streaming path continues. If the connector stops for any reason, upon restart it continues reading the WAL where it last left off. If it stops during a snapshot, it begins a new snapshot when it restarts.

The failure mode: a DDL change lands between the streaming path and the backfill path. The streaming path read the column under the old type. The backfill reads it under the new type. Both paths produce rows. Both paths pass row-count reconciliation. The values differ.

This is why reconciliation must run after backfills, not only after initial load. A backfill that re-reads a column after a type change can produce a different value than the streaming path for the same logical row. The reconciliation query is the only signal, because the connector will not tell you the type changed.

Iceberg’s schema evolution goals state that schema evolution supports safe column add, drop, reorder and rename, including in nested structures (Iceberg table spec). Safe in this context means the metadata is consistent. It does not mean your source and warehouse type declarations agree. The reconciliation query is still yours to write.

Pricing the fix

The query is cheap. Adding a singular dbt test or an Airflow task that compares SUM, scale probe, and per-row hash across source and warehouse is hours of work, not weeks. It runs on the same schedule as your existing reconciliation.

The expensive part is the type audit. You need to enumerate every money column on both sides, compare precision and scale declarations, and decide for each one whether to pin an explicit scale, migrate to DECFLOAT, or accept the rounding and document it. That audit is days to weeks depending on column count and how many teams own the schemas.

The migration decision is the expensive part. The query is the cheap part. Do the query first, because it tells you which columns actually need the migration decision.

Airflow’s TaskFlow API uses XComs to move inputs and outputs between tasks and requires that variables used as arguments be serializable (Airflow TaskFlow). If you pass reconciliation results between tasks, keep them to scalars and small dicts. Do not pass result sets through XCom.

The maintenance tax

Every new money column added to the warehouse is a new place the source/warehouse type contract can drift. The reconciliation query is a standing test, not a migration checklist item, because the drift is introduced by future schema changes, not by the original load.

The tax is not the query. The tax is the review step that asks, for every new money column, what the source scale is and what the warehouse scale is, and whether the difference is intentional. That review is minutes per column if it is part of the schema change process, and hours per column if it is discovered after the fact.

Row counts will keep matching. The money will keep not matching. The only thing that changes is whether you find out before or after the controller does.

FAQ

Does unique or not_null catch this? No. dbt ships with four generic data tests: unique, not_null, accepted_values, and relationships. None of them compares values across two systems. You need a singular test or a custom generic test that queries both sides.

Can I just compare SUM? No. SUM can hide offsetting errors. One row rounded up and another rounded down can net to the same total. Compare SUM, a scale probe, and a per-row hash in the same test.

Should I migrate everything to DECFLOAT? Not without pricing it. Snowflake notes that DECFLOAT may increase storage consumption and that NUMBER and FLOAT might provide better performance. Run the reconciliation query first to find which columns actually need exactness, then decide per column.

Why not just declare a scale on the source column? That is one valid fix. It makes the source coerce to the same scale as the warehouse, so the rounding happens on both sides. The tradeoff is that you are now rounding at the source, which may not be acceptable for the business logic that reads the source directly.

How often should the reconciliation run? After every backfill, after every schema change, and on the same schedule as your existing reconciliation. The backfill case is the one teams miss, because the backfill passes row-count reconciliation and looks healthy.

A PagerDuty schedule is an assignment mechanism. It says who gets the page. It does not say who owns the pipeline, what the page means, or what the on-call engineer is supposed to do when the page fires at 3 a.m. and the only person awake is the one holding the phone.

Data teams running Kafka, Airflow, dbt, Postgres/CDC, and Snowflake or Iceberg stacks tend to discover this gap the hard way. The schedule exists. The rotation exists. The ownership does not. When a consumer group rebalances, a source freshness check fails, or a Postgres statistics counter resets after a crash, the on-call engineer is left to reconstruct intent from dashboards and Slack threads.

This article is about what the rotation needs beyond the schedule. It is not a best-practices list. It is a set of load limits, escalation paths, and recovery procedures, each priced in incidents per shift, hours of follow-up, and the specific configuration keys that generate recurring operational work.

Start with a load budget, not a coverage chart

Google’s SRE book caps the amount of time SREs spend on purely operational work at 50%, with at least 50% allocated to engineering projects. The same chapter states that dealing with an on-call incident — root-cause analysis, remediation, and follow-up like writing a postmortem and fixing bugs — takes 6 hours on average, and that the maximum number of incidents per day is 2 per 12-hour on-call shift.

Those numbers are not universal. They are a load budget. The useful move for a data team is to derive its own budget from the same arithmetic: how many hours does a real incident consume, from the first page to the closed postmortem? If a Kafka consumer lag alert takes 90 minutes to triage and a backfill takes four hours to verify, then a shift that absorbs three of those has already spent the engineer’s follow-up capacity. The fourth page lands on someone who is already behind.

The SRE workbook restates the target: a maximum of two incidents per on-call shift, to ensure adequate time for follow-up. The workbook also notes that on-call engineers should be fully supported by procedures and escalation paths because being on-call can be daunting and highly stressful. That is not a wellness slogan. It is a statement about cognitive load. An engineer who is stressed and underslept makes worse decisions during an incident, and worse decisions during a data incident tend to mean a longer backfill or a wider blast radius.

For a data rotation, the load budget has to be measured in the units the team actually experiences: pages per shift, incidents per week, and hours of follow-up per incident. If the team cannot state those numbers, it does not have a rotation. It has a schedule.

Not every alert is a page

The single most common failure in a data on-call rotation is treating every alert as a page. A dbt source freshness failure, a Kafka consumer lag spike, and a Postgres statistics reset are not the same kind of event. They have different urgency, different recovery paths, and different follow-up work.

The dbt documentation makes the distinction concrete. dbt build does not include source freshness checks when building and testing resources in the DAG. If you select the Run source freshness checkbox in a job’s execution settings, dbt runs dbt source freshness as the first step and does not break subsequent steps if it fails. If you instead add dbt source freshness as a run step, and the source data is out of date, that step fails and subsequent steps do not run.

That is a configuration decision with an on-call consequence. The checkbox produces a non-breaking signal: the job continues, the freshness state is visible, and the on-call engineer can decide whether to act. The run step produces a hard stop: the pipeline halts, downstream models do not run, and the page fires. Neither is wrong. But the team has to decide which one it wants before the alert fires, not after.

The dbt documentation also recommends running source freshness jobs with at least double the frequency of the lowest SLA. A one-hour SLA implies a check every 30 minutes. A daily SLA implies a check every 12 hours. If the check frequency is wrong, the alert either fires too late to be useful or fires so often that the on-call engineer learns to ignore it.

There is a further limitation worth knowing. dbt source freshness for Snowflake is calculated using the LAST_ALTERED column. That column reflects metadata changes, not necessarily data changes. A table can be altered without new rows arriving, and a table can receive new rows without the metadata changing in the way the check expects. The check is a signal, not a proof.

Kafka consumer lag has its own timing semantics. The Confluent consumer configuration reference documents heartbeat.interval.ms as the expected time between heartbeats to the consumer coordinator when using group management, with a default of 3000 ms and a note that it should typically be no higher than one-third of session.timeout.ms. It documents session.timeout.ms as the timeout used to detect client failures, with a default of 45000 ms. It documents max.poll.interval.ms as the maximum delay between invocations of poll(), with a default of 300000 ms, after which the consumer is considered failed and the group rebalances.

Those three values interact. A consumer that processes a batch slowly can exceed max.poll.interval.ms and be evicted from the group even though it is healthy. A consumer that is paused for a deploy can exceed session.timeout.ms and trigger a rebalance. The alert that fires is “consumer lag,” but the cause may be a configuration value, not a data volume problem. The on-call engineer needs to know which one before touching anything.

Escalation paths and named owners

Google’s SRE book lists clear escalation paths, well-defined incident-management procedures, and a blameless postmortem culture as the most important on-call resources. It also describes primary and secondary on-call rotations, with duties varying by team: the secondary may be a fall-through for pages the primary misses, or may handle non-urgent production activities while the primary handles pages.

Data teams need the same structure, but the ownership map is harder. A pipeline may be owned by a data engineer, an analytics engineer, or a platform engineer. The Kafka cluster may be owned by a platform team. The warehouse may be owned by a separate group. When a page fires, the on-call engineer needs to know which of those owners to escalate to, and what response time to expect.

That information has to be written down before the rotation starts. A useful artifact is a one-page ownership table per pipeline: pipeline name, primary owner, secondary owner, escalation contact, expected response time, and the specific failure modes that justify a page. The table is not documentation for its own sake. It is the difference between a 10-minute escalation and a 40-minute Slack search at 3 a.m.

The SRE workbook describes playbooks as high-level instructions on how to respond to automated alerts, explaining severity and impact, and including debugging suggestions and possible actions. It also recommends implementing automation if playbooks are a deterministic list of commands the on-call engineer runs every time a particular alert fires. That recommendation is directly applicable to data pipelines. If the playbook for a freshness failure is “run this query, check this table, restart this task,” the playbook should be a script, not a document.

The maintenance tax in configuration

The recurring operational work in a data stack does not come from the big architectural decisions. It comes from configuration drift. Three examples, each anchored to a documented behavior.

Kafka consumer timeouts. The defaults for heartbeat.interval.ms, session.timeout.ms, and max.poll.interval.ms are tuned for general-purpose consumers. A consumer that does heavy per-record processing, or that pauses during a deploy, may need different values. Every change to those values is a change to the failure mode. The on-call engineer needs to know which consumers have non-default values and why.

Postgres statistics collection. PostgreSQL’s cumulative statistics system supports collection and reporting of information about server activity, including accesses to tables and indexes in disk-block and individual-row terms. Collection is controlled by parameters such as track_activities, track_counts, track_functions, and track_io_timing. The statistics views do not update instantaneously: each server process flushes accumulated statistics to shared memory just before going idle, but not more frequently than once per PGSTAT_MIN_INTERVAL milliseconds, so the displayed information lags behind actual activity. And when a server starts from an unclean shutdown — after an immediate shutdown, a server crash, a base backup, or point-in-time recovery — all statistics counters are reset.

That last point matters for on-call. A dashboard that depends on cumulative counters will show a discontinuity after a crash. An alert threshold based on a counter that just reset will either fire spuriously or fail to fire. The on-call engineer needs to know which dashboards are counter-based and which are gauge-based, and what happens to each after a restart.

dbt freshness check placement. As noted above, the checkbox and the run step produce different failure behavior. The choice is a maintenance decision. If the check is a run step, every freshness failure is a pipeline halt and a page. If the check is a checkbox, every freshness failure is a signal that someone has to review. The first option creates more pages. The second creates more silent failures. The team has to pick which tax it wants to pay.

A rotation checklist that fits on one page

Before anyone goes on-call for a data pipeline, the following should exist and be current. This is not a maturity model. It is the minimum set of artifacts that makes the rotation functional.

• Ownership table. Pipeline name, primary owner, secondary owner, escalation contact, expected response time.
• Alert classification. Which alerts page, which alerts create a ticket, and which alerts are informational. For each paging alert, the specific failure mode it represents.
• Load budget. The team’s target for incidents per shift and hours of follow-up per incident, derived from its own incident history.
• Playbook per paging alert. Severity, impact, debugging steps, and the actions that mitigate or resolve the alert. If the steps are deterministic, they should be a script.
• Recovery procedures. How to restart a consumer group, how to re-run a failed dbt model, how to verify a backfill, and how to confirm that a schema change did not break a downstream consumer.
• Handoff template. What the outgoing on-call engineer writes down: open incidents, in-progress backfills, known configuration changes, and anything that is likely to page in the next shift.

The SRE workbook describes a training approach that is worth adapting: a checklist of focus areas, lab sessions for common debugging and mitigation tasks, and “Wheel of Misfortune” exercises where the team role-plays recent incidents. For a data team, the equivalent is a game day that exercises the actual failure modes: a backfill that runs long, a schema change that breaks a consumer, a freshness check that fails silently, and an escalation that reaches the wrong person.

The game day is not a drill for its own sake. It is how the team discovers which parts of the rotation are missing before the pager discovers it for them.

What the schedule cannot do

A PagerDuty schedule answers one question: who is holding the phone. It does not answer who owns the pipeline, what the page means, how long the recovery should take, or who to call when the first three steps do not work.

Those answers come from a load budget measured in incidents per shift, an alert classification that separates pages from signals, an ownership table with named escalation contacts, and recovery procedures that have been tested. The schedule is necessary. It is not sufficient.

The test is simple. Ask the on-call engineer to describe, without looking anything up, what happens when a Kafka consumer group rebalances at 3 a.m. If the answer is a specific sequence of checks, a named escalation contact, and a known recovery procedure, the rotation is working. If the answer is “I would look at the dashboard and figure it out,” the schedule is doing all the work, and the pipeline has no owner.

FAQ

How many incidents per shift should a data on-call rotation target?

Google’s SRE workbook targets a maximum of two incidents per on-call shift to allow adequate time for follow-up. That number is a starting point, not a universal rule. A data team should derive its own target from the average time a real incident consumes, including triage, remediation, and postmortem. If a typical incident takes four hours, two incidents per shift already exceeds a normal working day.

Should dbt source freshness checks break the pipeline or just warn?

It depends on the SLA and the downstream dependency. The dbt documentation describes two behaviors: the Run source freshness checkbox runs the check as a non-breaking first step, while adding dbt source freshness as a run step causes the step to fail and subsequent steps not to run. The first produces a signal; the second produces a halt. The team should choose based on whether downstream models can tolerate stale source data.

Why does a Kafka consumer get evicted from its group even when it is healthy?

The Confluent consumer configuration reference documents max.poll.interval.ms as the maximum delay between invocations of poll(), with a default of 300000 ms. If the consumer’s processing loop takes longer than that between polls, the consumer is considered failed and the group rebalances. A consumer that processes large batches or pauses during a deploy can hit this limit without any underlying data problem.

What happens to Postgres monitoring after a crash?

The PostgreSQL documentation states that when a server starts from an unclean shutdown — after an immediate shutdown, a server crash, a base backup, or point-in-time recovery — all statistics counters are reset. Dashboards and alerts that depend on cumulative counters will show a discontinuity. The on-call engineer should know which monitoring depends on those counters and how the alert thresholds behave after a reset.

Do we need a secondary on-call rotation for data pipelines?

Google’s SRE book describes primary and secondary rotations with duties that vary by team. For a data team, the secondary serves two purposes: fall-through for pages the primary misses, and a second person who knows the recovery procedures. The second purpose is the more important one. If only one person knows how to recover a pipeline, the rotation has a single point of failure that the schedule does not address.

A producer team renames user_id to account_id in an Avro record. They run the schema through Schema Registry. It is accepted. They post in the shared channel: “Non-breaking change, no consumer action needed.”

Two days later, a consumer job fails to deserialize. A dbt incremental model silently stops populating a column. A CDC pipeline replays a batch it already processed.

The rename was not non-breaking. It was non-breaking relative to the compatibility mode that was actually in effect, the schema format in use, and the set of consumers that actually read the topic. The producer team checked one of those three things.

This is not a story about a careless team. It is a story about a review process that treats “the registry accepted it” as equivalent to “nothing downstream will break.” Those are different claims, and the gap between them is where on-call hours go to die.

What the registry actually checks

Schema Registry enforces compatibility by comparing a new schema version against previous versions using a configurable compatibility type. The default is BACKWARD, not BACKWARD_TRANSITIVE. That distinction matters more than most teams realize.

Under BACKWARD, a consumer using the new schema can process data written by producers using schema X or X-1, but not necessarily X-2. Under BACKWARD_TRANSITIVE, that same consumer can process data written by X, X-1, or X-2. The Confluent documentation is explicit: the default is BACKWARD, and the main reason is so that you can rewind consumers to the beginning of the topic.

Here is the failure mode. A team has three schema versions in production. They add a fourth. The registry checks version 4 against version 3 under BACKWARD. It passes. But a consumer that rewinds to the beginning of the topic will encounter version 1 and version 2 messages. If the change is not transitive-safe, that consumer breaks on old data.

The registry did its job. The review did not.

What each compatibility type actually permits

The Confluent compatibility tables for Avro and Protobuf show which operations are allowed under each mode. Adding an optional field is compatible under BACKWARD, FORWARD, and FULL. Removing an optional field is also compatible under all three. Adding a required field is compatible only under FORWARD. Removing a required field is compatible only under BACKWARD.

Renames are not listed as a distinct operation. In Avro, a rename is typically expressed as a field removal plus a field addition. Whether that passes depends on whether the removed field was optional or had a default value, and whether the added field has a default value. The documentation states: “the ability to delete a field and keep the schema compatible requires that the field was either specified as optional or provided a default value in the original version.”

So a rename can pass BACKWARD if the old field had a default and the new field has a default. It can pass FORWARD under similar conditions. It can pass FULL if both conditions hold. But passing the registry check does not mean consumers will find the data they expect. A consumer looking for user_id will not find it in a message that only contains account_id. The registry does not know what field names your consumer code references.

Schema format changes the rules

Avro, Protobuf, and JSON Schema have different compatibility rules. The Confluent documentation notes that Avro was developed with schema evolution in mind and its specification clearly states the rules for backward compatibility, whereas the rules for JSON Schema and Protobuf can be more nuanced.

For JSON Schema, compatibility behavior depends on both the compatibility policy (lenient or strict) and the content model (additionalProperties: true for open, false for closed). A change that passes under a lenient policy with an open content model may fail under a strict policy with a closed content model. The review must confirm which policy and which content model are actually in effect.

For Protobuf, the documentation notes that best practice is to use BACKWARD_TRANSITIVE, because adding new message types is not forward compatible. A team using BACKWARD with Protobuf may accept a change that breaks forward compatibility in ways the registry does not flag.

The effective compatibility mode may not be what you think

A REST API call to compatibility mode is global and overrides any compatibility parameters set in schema registry properties files. This means the effective mode for a subject may differ from what the properties file says. A team that set BACKWARD_TRANSITIVE in their properties file may find that a global API call reset it to BACKWARD.

The review must verify the effective mode for the specific subject, not the mode someone believes is configured. The diagnostic is straightforward:

curl -s http://schema-registry:8081/config
curl -s http://schema-registry:8081/config/<subject-name>

The first call returns the global compatibility level. The second returns the subject-level override, if any. If the subject-level value is absent, the global value applies. If a global API call was made, it overrides the properties file.

What the registry does not check

The registry checks schema compatibility. It does not check:

• Whether consumers have been rewound to the beginning of the topic
• Whether dbt incremental models will pick up the change
• Whether downstream CDC pipelines handle the renamed field
• Whether any consumer code references the old field name
• Whether the change is transitive-safe across all schema versions in the topic

Each of these is a separate failure mode. Each requires a separate check.

The dbt layer: silent column drops

dbt incremental models have an on_schema_change configuration. The default is ignore. Under ignore, if you add a column to your incremental model and execute a dbt run, the column will not appear in the target table. If you remove a column and execute a dbt run, dbt will fail.

This means a renamed field can produce two different failure modes depending on which side of the rename the dbt model sees. If the model references the old field name and the source no longer provides it, the run fails. If the model references the new field name and the source provides it, but the target table was built with the old schema, the new column silently does not appear.

The documentation is explicit: “None of the on_schema_change behaviors backfill values in old records for newly added columns.” If you need to populate those values, you must run manual updates or trigger a --full-refresh.

There is another constraint: on_schema_change only tracks top-level column changes. It does not track nested column changes. A rename inside a nested structure will not trigger a schema change, even if on_schema_change is set appropriately.

The diagnostic is to check which models use incremental materialization and what their on_schema_change setting is:

dbt ls -s config.materialized:incremental --output json | jq '.[].config.on_schema_change'

If the output is null or "ignore", the model will not pick up new columns automatically.

The CDC layer: replay and idempotency

PostgreSQL logical decoding slots emit each change once in normal operation. But the current position of each slot is persisted only at checkpoint. In the case of a crash, the slot might return to an earlier LSN, which will cause recent changes to be sent again when the server restarts.

The documentation states: “Logical decoding clients are responsible for avoiding ill effects from handling the same message more than once.”

This means a CDC pipeline that consumes a renamed field must be idempotent against replay. If the pipeline processes a message with user_id, then a message with account_id, then a replayed message with user_id, it must not produce duplicate or inconsistent rows.

Replication slots persist across crashes and know nothing about the state of their consumers. They will prevent removal of required resources even when there is no connection using them. A slot that is no longer required should be dropped, but dropping it requires knowing which consumers depend on it.

The diagnostic is to check which slots exist and how far behind they are:

SELECT slot_name, plugin, slot_type, active, restart_lsn,
       pg_size_pretty(pg_wal_lsn_diff(pg_current_wal_lsn(), restart_lsn)) AS lag
FROM pg_replication_slots;

A slot with a large lag and no active connection is a candidate for investigation. It may be holding WAL for a consumer that no longer exists, or it may be a consumer that is about to replay a large batch.

The Iceberg layer: column IDs and rename safety

The Iceberg spec states that schema evolution supports safe column add, drop, reorder and rename, including in nested structures. This is a stronger guarantee than what Schema Registry provides for Kafka topics, because Iceberg tracks column IDs rather than column names.

But the spec also notes that the format version number is incremented when new features are added that will break forward-compatibility. This means a table written with format version 3 may not be readable by a reader that only supports version 2. The review must confirm which format version the table uses and whether all readers support it.

The diagnostic is to check the table metadata:

SELECT * FROM catalog.db.table_snapshots LIMIT 1;
-- or
SELECT * FROM "table$snapshots" LIMIT 1;

The format-version field in the table metadata indicates which version the table uses. If the table was upgraded from version 2 to version 3, readers that only support version 2 will fail.

What a cross-team schema review must actually check

The review is not a single check. It is a sequence of checks, each of which can fail independently.

1. Confirm the effective compatibility mode for the specific subject. Use the Schema Registry API to check both the global and subject-level settings. Do not rely on the properties file.
2. Confirm the schema format and its specific rules. Avro, Protobuf, and JSON Schema have different compatibility rules. JSON Schema compatibility depends on both the policy and the content model.
3. Confirm whether the change is transitive-safe. If any consumer rewinds to the beginning of the topic, the change must be compatible with all schema versions, not just the last one.
4. Confirm whether any consumer rewinds to the beginning of the topic. This is a property of the consumer configuration, not the schema. Check consumer group offsets and retention settings.
5. Confirm whether downstream dbt models use on_schema_change and what setting. The default is ignore, which means new columns silently do not appear.
6. Confirm whether CDC pipelines are idempotent against replay. Logical decoding slots can replay changes after a crash. The pipeline must handle duplicate messages.
7. Confirm whether the change is compatible with the Iceberg table format version in use. A table upgraded to a newer format version may not be readable by older readers.

Pricing the maintenance tax

The cost of a schema change is not paid at the moment of the change. It is paid when a consumer fails to deserialize, when a dbt incremental model silently drops a column, or when a CDC pipeline replays a change it already processed.

Each of these failure modes has a different cost profile:

• Consumer deserialization failure: The consumer stops processing. If it is a real-time pipeline, the lag grows. If it is a batch pipeline, the batch fails. The fix is to update the consumer code and redeploy. The cost is the time to diagnose, fix, and redeploy, plus the cost of any data that was not processed during the outage.
• dbt silent column drop: The model runs successfully but produces incomplete data. The failure is not detected until someone notices that a column is null or missing. The fix is to run a full refresh, which may be expensive if the model processes a large volume of data. The cost is the compute cost of the full refresh plus the time to diagnose why the column disappeared.
• CDC replay: The pipeline processes a message it already processed. If the pipeline is not idempotent, it produces duplicate rows. The fix is to deduplicate the data and make the pipeline idempotent. The cost is the time to diagnose the duplication plus the cost of the deduplication job.

The review should price these failure modes in on-call hours, not just in registry API calls. A review that takes 30 minutes and catches a non-transitive change is cheaper than a review that takes 5 minutes and misses it.

Frequently asked questions

Does Schema Registry check for field renames?

Schema Registry checks compatibility based on the rules for the schema format and compatibility type. In Avro, a rename is typically expressed as a field removal plus a field addition. Whether that passes depends on whether the removed field was optional or had a default value, and whether the added field has a default value. The registry does not know what field names your consumer code references.

What is the difference between BACKWARD and BACKWARD_TRANSITIVE?

Under BACKWARD, a consumer using the new schema can process data written by producers using schema X or X-1, but not necessarily X-2. Under BACKWARD_TRANSITIVE, that same consumer can process data written by X, X-1, or X-2. The default is BACKWARD.

Why does my dbt incremental model not pick up a new column?

The default on_schema_change setting is ignore. Under ignore, if you add a column to your incremental model and execute a dbt run, the column will not appear in the target table. You must set on_schema_change to append_new_columns or sync_all_columns, or run a full refresh.

Can a CDC pipeline process the same message twice?

Yes. PostgreSQL logical decoding slots persist their position only at checkpoint. In the case of a crash, the slot might return to an earlier LSN, which will cause recent changes to be sent again when the server restarts. Logical decoding clients are responsible for avoiding ill effects from handling the same message more than once.

Does Iceberg handle column renames safely?

The Iceberg spec states that schema evolution supports safe column add, drop, reorder and rename, including in nested structures. This is because Iceberg tracks column IDs rather than column names. However, the format version number is incremented when new features are added that will break forward-compatibility, so readers must support the table’s format version.

Sources

• Confluent Schema Registry documentation, “Schema Evolution and Compatibility”: https://docs.confluent.io/platform/current/schema-registry/fundamentals/schema-evolution.html
• dbt documentation, “Incremental models”: https://docs.getdbt.com/docs/build/incremental-models
• PostgreSQL documentation, “Logical Decoding Concepts”: https://www.postgresql.org/docs/current/logicaldecoding-explanation.html
• Apache Iceberg documentation, “Iceberg Table Spec”: https://iceberg.apache.org/spec/
• Apache Airflow documentation, “Dynamic Task Mapping”: https://airflow.apache.org/docs/apache-airflow/stable/authoring-and-scheduling/dynamic-task-mapping.html

Twelve million Parquet files at 3 MB each is 36 TB of data. It is also a table that will make your query planner sweat before it reads a single row. The failure mode is not a crash. It is a slow, expensive drift: planning times climb, manifest lists grow, and every compaction job you run seems to cost more than the last one.

This article prices rewrite_data_files for that shape of table and gives you a diagnostic you can run on your own catalog before you schedule anything.

What the procedure actually does

Iceberg’s rewrite_data_files is a Spark action that reads groups of small data files and writes them back as larger files. The Iceberg maintenance documentation describes it as combining small files into larger files to reduce metadata overhead and runtime file open cost. It is not a metadata-only operation. It reads Parquet bytes and writes new Parquet bytes.

The procedure exposes options that control how much it reads and writes in one run. The documented options include target-file-size-bytes, min-input-files, max-concurrent-file-group-rewrites, partial-progress, and delete-file-threshold. The Iceberg docs show target-file-size-bytes being set explicitly in the Spark action example, and the Snowflake Iceberg documentation points at delete-file-threshold and rewrite-all as the options that matter when position deletes accumulate.

Two of those options decide your bill.

min-input-files sets the smallest group of files the procedure will consider worth rewriting. If you leave it low, the procedure will happily rewrite pairs of 3 MB files into 6 MB files. That is a bad trade. You paid a full read and write to move from 12 million files to 6 million files, and you still have a small-file problem.

target-file-size-bytes sets the size of the output. If you set it to 512 MB, each rewrite group produces roughly one 512 MB file per partition. That is the number that determines how many files you have after the job, and therefore how many times you need to run the job again.

partial-progress controls whether the procedure commits work as it goes or holds everything until the end. Partial progress reduces the blast radius of a failed run, but it also produces more snapshots and more commits. Each commit is a metadata write. On a table with millions of files, metadata writes are not free.

The arithmetic of 12 million files

Assume 36 TB of data in 12 million files, average 3 MB. Assume you set target-file-size-bytes to 512 MB and min-input-files high enough that the procedure only touches groups that will produce at least one full target file.

At 512 MB per output file, 36 TB becomes roughly 72,000 files. That is a 166x reduction in file count. It is also a full read of 36 TB and a full write of 36 TB.

On object storage, that is 72 TB of I/O. At S3 standard pricing, data transfer out is not the issue for same-region reads, but request costs are. A 3 MB file read is one GET. Twelve million GETs is twelve million requests. S3 charges per 1,000 requests, so the request cost alone is in the low hundreds of dollars for the reads, plus the PUTs for the new files. The dominant cost is not the request line item. It is the compute hours spent reading and writing 72 TB.

If your Spark cluster sustains 2 GB/s of aggregate read throughput, 36 TB takes about 5 hours. If it sustains 500 MB/s, it takes about 20 hours. The write side is usually slower because Parquet encoding and compression are CPU-bound. A realistic single full rewrite of this table is a multi-hour job on a cluster large enough to matter.

That is the number to put in front of whoever approves the cluster budget. It is not a nightly job. It is a quarterly job at best, and only if the write pattern that created the small files has been fixed.

Why the small files exist

Before you schedule a rewrite, find out what is producing 3 MB files. The Iceberg maintenance docs are explicit that streaming queries may produce small data files that should be compacted. If your ingestion is a Kafka Connect sink or a Flink job committing every minute, you will get small files by design. Rewriting them without changing the commit interval is a treadmill.

The diagnostic is the files metadata table. The Iceberg docs call it useful for inspecting data file sizes and determining when to compact partitions. In Athena, the equivalent is SELECT * FROM "dbname"."tablename$files". In Trino, the Iceberg connector exposes metadata tables you can query directly.

Run this against your own table and look at the distribution, not the average:

SELECT
  partition,
  count(*) AS file_count,
  sum(file_size_in_bytes) / 1024 / 1024 AS total_mb,
  avg(file_size_in_bytes) / 1024 / 1024 AS avg_mb,
  min(file_size_in_bytes) / 1024 / 1024 AS min_mb,
  max(file_size_in_bytes) / 1024 / 1024 AS max_mb
FROM my_table.files
GROUP BY partition
ORDER BY file_count DESC
LIMIT 20;

If the top partitions have hundreds of thousands of files and the bottom partitions have a handful, you do not need a full-table rewrite. You need a partition-filtered rewrite. The Iceberg Spark action supports a filter expression, and the docs show it being used with an equality predicate on a date column. That is the difference between a 20-hour job and a 20-minute job.

Manifests are the other half of the cost

File count does not only affect data reads. The Iceberg spec describes the metadata tree as an index over the table’s data. Manifests track data files, and manifest lists track manifests. The spec’s stated goal is O(1) remote calls to plan a scan, not O(n) where n grows with the number of files.

That goal holds only if manifests are reasonably sized. The Iceberg maintenance docs note that more data files leads to more metadata stored in manifest files, and that small data files cause an unnecessary amount of metadata and less efficient queries from file open costs. They also document rewriteManifests as a separate operation for regrouping data files into manifests when the write pattern does not align with the query pattern.

This matters for your rewrite budget because rewrite_data_files does not fix manifest layout. It produces new data files, and those new files get recorded in new manifests. If your existing manifests are already fragmented, you may need rewriteManifests after the data rewrite. That is a second job with its own runtime.

The docs show rewriteManifests being called with a rewriteIf predicate on manifest file length, for example rewriting manifests smaller than 10 MB. That is a cheap operation compared to rewriting data, but it is not zero. It reads and writes manifest files, and on a table with millions of files, there are a lot of manifests.

Delete files change the calculus

If your table uses merge-on-read deletes, rewrite_data_files has a second job: applying delete files to data files. The Iceberg spec defines two delete types. Position deletes mark a row deleted by data file path and row position. Equality deletes mark a row deleted by column values.

The Snowflake Iceberg documentation is direct about the operational consequence: excessive position deletes, especially dangling position deletes, might prevent table creation and refresh operations, and the recommended fix is table maintenance using rewrite_data_files with delete-file-threshold or rewrite-all.

That means your rewrite frequency is not just a function of file size. It is a function of delete volume. A table with heavy update traffic accumulates position deletes faster than it accumulates small files. If you set delete-file-threshold too high, readers pay the merge cost on every query. If you set it too low, you rewrite data files that have only a few deletes, which is wasted I/O.

The diagnostic is the delete file count per data file. Query the metadata tables for delete file counts and compare them to data file counts. If the ratio is climbing, your rewrite cadence needs to be tied to delete accumulation, not to a calendar.

What it costs in hours and dollars

Here is a priced example. These are illustrative numbers based on the arithmetic above, not benchmarks from a specific cluster. Substitute your own throughput and instance pricing.

A full rewrite of 36 TB at 2 GB/s aggregate read and 1 GB/s aggregate write takes roughly 5 hours of read time and 10 hours of write time, overlapped to about 10 hours wall clock. A 100-node cluster of 16 vCPU instances at $0.50 per instance-hour costs $500 for the run. Add S3 request costs in the low hundreds. Call it $700 to $900 per full rewrite.

If you run that monthly, it is $8,400 to $10,800 per year. If you run it weekly, it is $36,000 to $47,000 per year. The difference between monthly and weekly is not a best-practice question. It is a budget question, and the answer depends on how fast your write pattern recreates the problem.

The cheaper path is to fix the write side. If you can raise the commit interval on the streaming job that produces 3 MB files, you reduce the rate at which small files appear. That is a config change, not a cluster run. The Iceberg docs describe the problem as streaming queries producing small data files. The fix is usually in the writer, not in the maintenance job.

How often to run it

There is no universal cadence. There is a decision procedure.

First, measure the file count and the average file size per partition. If the average is below your target and the count is growing, you have a compaction need.

Second, measure the query planning time. If planning is a small fraction of total query time, compaction is not urgent. If planning dominates, it is.

Third, measure the delete file ratio. If deletes are accumulating faster than files, your cadence is set by deletes.

Fourth, price the rewrite. If the rewrite costs more than the query slowdown it prevents, do not run it. That is the tradeoff. A table with 12 million files that is queried twice a day may not justify a $900 monthly rewrite. A table with 12 million files that is queried continuously by a dashboard may justify it weekly.

The Iceberg docs recommend regularly expiring snapshots and cleaning metadata files, and they note that tables with frequent commits may need to regularly clean metadata files. That is a separate maintenance tax from data compaction. Budget for both.

What to run before you schedule anything

Run the file distribution query above. Run a query against the manifests metadata table to see manifest count and sizes. Run a query against the snapshots table to see how many snapshots you are retaining. Check write.metadata.previous-versions-max and write.metadata.delete-after-commit.enabled in your table properties. The Iceberg docs give the defaults as 100 and false, and they show that with deletion disabled, orphaned metadata files accumulate and can only be cleaned with orphan file deletion.

Then run a partition-filtered rewrite on your worst partition and time it. Multiply by the number of partitions you actually need to fix. That is your real cost. It is almost always less than a full-table rewrite, and it is almost always more than the estimate you gave your manager.

FAQ

Does rewrite_data_files rewrite the whole table by default?

No. It operates on file groups selected by the procedure’s options. You can filter by partition. The Iceberg Spark action supports a filter expression, and the docs show it being used with an equality predicate.

What is a safe target-file-size-bytes?

The Iceberg docs show 500 MB in their example. Trino’s Iceberg connector defaults iceberg.target-max-file-size to 1 GB. The right value depends on your query engine’s split size and your object store’s read characteristics. Larger files reduce metadata overhead but increase the cost of rewriting any single file.

Can I run it while writes are happening?

Iceberg uses optimistic concurrency. The spec says a change that rewrites files can be applied to a new table snapshot if all of the rewritten files are still in the table. If a concurrent write removes or replaces a file in your rewrite group, the commit may fail and retry. On a busy table, expect retries.

Do I need rewriteManifests too?

Only if your manifest layout is misaligned with your query pattern. The Iceberg docs describe rewriteManifests as regrouping data files into manifests when the write pattern does not align with the query pattern. It is a separate operation from data compaction.

How do I know if delete files are the problem?

Query the metadata tables for delete file counts. The Snowflake Iceberg docs warn that excessive position deletes can prevent table creation and refresh, and recommend rewrite_data_files with delete-file-threshold or rewrite-all. If your delete file count is climbing faster than your data file count, compaction cadence should follow deletes.

Sources

Apache Iceberg maintenance documentation: https://iceberg.apache.org/docs/latest/maintenance/

Apache Iceberg Spark procedures documentation: https://iceberg.apache.org/docs/latest/spark-procedures/

Apache Iceberg table specification: https://iceberg.apache.org/spec/

Amazon Athena Iceberg table documentation: https://docs.aws.amazon.com/athena/latest/ug/querying-iceberg-table-data.html

Trino Iceberg connector documentation: https://trino.io/docs/current/connector/iceberg.html

Snowflake Iceberg tables documentation: https://docs.snowflake.com/en/user-guide/tables-iceberg

Eighteen months after a data mesh rollout, the platform team’s ticket queue is the only honest scoreboard. Not the architecture diagram. Not the governance charter. The queue.

This piece is about counting those tickets. Not to argue data mesh is wrong — Dehghani’s four principles are coherent and the central-team bottleneck is real — but to price what domain ownership actually shifts onto domain teams and what it leaves behind on the platform team’s on-call rotation.

What the principles actually assign

Dehghani’s original article names four principles: domain-oriented decentralized data ownership, data as a product, self-serve data infrastructure as a platform, and federated computational governance. Each principle maps to a concrete operational responsibility, and each responsibility generates a ticket category.

Domain ownership moves analytical data responsibility to the team closest to the source. Data as a product means that team owes consumers a contract — schema, semantics, quality attributes, SLOs. Self-serve platform means the platform team provides domain-agnostic tooling. Federated governance means policies are agreed across domains, not handed down.

None of those four principles say who fixes a broken Debezium connector at 02:00. That gap is where the tickets live.

Ticket taxonomy: what actually arrived

Across an eighteen-month window, platform tickets tend to fall into six buckets. The counts vary by org, but the categories are stable.

1. Access and entitlement. Domain teams need read access to upstream data products. Federated governance says the domain owns the grant. In practice, the platform team owns the IAM plumbing, so every grant becomes a ticket.
2. Pipeline failures on shared infrastructure. A domain’s dbt model fails because an upstream source changed. The domain owns the model. The platform owns the scheduler, the warehouse, and the connection pool. The ticket lands on the platform queue.
3. Schema change coordination. A producer domain renames a column. Consumers break. The contract said this would be governed. The governance group meets biweekly.
4. Data quality incidents. A dbt not_null or relationships test fails. The domain owns the test. The platform owns the alerting route and the on-call rotation.
5. Backfill and CDC recovery. A Kafka topic is reset, or a Postgres logical replication slot falls behind, or an Iceberg snapshot expires before a consumer reads it. Someone has to decide who replays what.
6. Cost and performance. A domain’s Snowflake query scans too many micro-partitions. The domain owns the model. The platform owns the credit budget.

Buckets 1, 2, 4, and 6 are the ones that quietly grow. They are not new work in the sense that the work did not exist before. They are reclassified work: tasks the central team used to do silently, now surfaced as tickets because the ownership boundary is explicit.

The reclassification problem

This is the single most important measurement issue. If you count tickets before and after a data mesh rollout and conclude that platform load increased, you may be measuring a change in visibility, not a change in work.

Before the mesh, a central data engineer fixed a broken pipeline, updated a schema, and re-ran a backfill without opening a ticket. After the mesh, the same fix requires a domain team to request platform action, which opens a ticket. The work is identical. The ticket count is not.

To separate new work from reclassified work, you need a counterfactual. The cleanest one is a pre-rollout baseline of platform-team hours by activity, not ticket count. If you did not capture that baseline, you cannot honestly claim the mesh increased load. You can only claim it changed the shape of the queue.

What the platform team actually owns after the mesh

The self-serve platform principle is the one most often under-specified. Dehghani’s article describes a multi-plane platform: a control plane for policy and a data plane for compute and storage. The platform team owns the control plane. Domain teams own their data products on the data plane.

In practice, the control plane is where the tickets concentrate. Access control, schema registry, catalog metadata, lineage, alerting routes, and cost attribution all live in the control plane. Every domain team that onboards adds a row to each of those systems, and every row is a potential ticket.

The platform team’s on-call load does not shrink when domain teams take ownership of their pipelines. It shifts from “fix the pipeline” to “fix the platform primitive that the pipeline depends on.” That is a different skill set and a different pager.

Schema evolution: the mechanics that become domain-team work

Schema evolution is where the ownership boundary is most visible and most expensive.

Apache Iceberg’s spec supports safe column add, drop, reorder, and rename, including in nested structures. That is a format-level guarantee. It does not tell you which consumer will break when a producer renames a column. The format allows the change. The contract governs the change. The domain team executes the change.

If the domain team has never run a schema evolution against a live consumer, the first one is expensive. The failure mode is not the rename itself. It is the downstream dbt model that selects column_name and now returns nulls, or the Kafka consumer that deserializes against an old Avro schema and drops the record.

dbt’s data tests are the cheapest guardrail here. A not_null test on a column that should never be null will fail the moment a rename lands. A relationships test will fail when a foreign key stops resolving. Both are SQL queries that return failing rows; if the query returns zero rows, the assertion passes. That is the entire mechanism. It is not sophisticated. It is also the difference between catching a schema break in the PR and catching it in a dashboard three days later.

The ticket that follows a missed schema break is not a schema ticket. It is a data quality ticket, a backfill ticket, and a trust ticket, in that order.

CDC and backfill: the recovery path that nobody owns

Change data capture is the second place where the ownership boundary frays.

Debezium reads the Postgres write-ahead log through a logical replication slot. If the slot falls behind — because the consumer stalled, because the connector restarted, because the network partition lasted longer than the WAL retention — the slot is dropped and the change stream has a gap. Recovering that gap requires a backfill from the source table, which requires a snapshot, which requires coordination with the source team, which is a different domain.

Who owns that ticket? The producer domain owns the source table. The consumer domain owns the downstream model. The platform team owns the connector. Three teams, one gap, and no single owner.

The same pattern appears with Iceberg. Snapshots are retained per the table’s snapshot retention policy. If a consumer reads a snapshot that has expired, the read fails. The fix is a re-read from a newer snapshot, which may not be equivalent if the consumer was doing incremental processing. That is a backfill, and it is a domain-team responsibility that the domain team may not have known it signed up for.

Airflow: where dynamic task mapping hides the cost

Dynamic task mapping is a common pattern in domain-owned pipelines. A task generates a list at runtime, and the scheduler creates one task instance per element. The number of task instances is not known at DAG parse time.

This is useful. It is also a ticket generator. When a mapped task fails, the failure is per-instance, and the retry semantics depend on whether the mapping was task-generated or static. Airflow’s documentation notes that task-generated mapping cannot be used with TriggerRule.ALWAYS, because the expanded parameters are undefined at the time the task would execute. That constraint is enforced at DAG parse time.

The operational consequence: a domain team that adopts dynamic task mapping without understanding the trigger rule constraint will hit a parse error, open a ticket, and wait. The fix is a one-line change. The ticket is not.

Snowflake: the clustering key that becomes a platform ticket

Clustering keys are the clearest example of a domain-owned decision that generates platform-owned cost.

Snowflake’s documentation is explicit: clustering keys are not intended for all tables, because of the cost of initially clustering the data and maintaining the clustering. Reclustering consumes credits and generates new micro-partitions. The original micro-partitions are retained for Time Travel and Fail-safe, which means storage costs increase.

A domain team that adds a clustering key to a large table to fix a slow query is making a cost decision on the platform team’s budget. If the table has high DML volume, the reclustering cost is ongoing. The domain team sees a faster query. The platform team sees a credit line item.

The ticket that follows is not a clustering ticket. It is a cost review ticket, and it arrives at the end of the month, when the domain team has moved on.

Pricing the maintenance tax

Here is a reproducible way to estimate the per-domain maintenance tax without inventing numbers.

1. Export the platform ticket queue for the last six months. Group by domain and by category using the six buckets above.
2. For each ticket, record the time from open to close, and the number of distinct people who touched it.
3. Separate tickets that required a code change from tickets that required only a configuration change or an access grant.
4. Multiply configuration and access tickets by the average handling time. That is the coordination tax.
5. Multiply code-change tickets by the average handling time plus the average review time. That is the engineering tax.
6. Add the platform team’s on-call hours attributable to domain-owned pipelines. That is the pager tax.

The sum is the monthly maintenance tax per domain. It is not a research finding. It is an arithmetic exercise on your own queue. The number will be different in every org, and that is the point.

What the eighteen-month count actually shows

Three patterns tend to hold across orgs that run this exercise honestly.

First, the platform team’s ticket volume does not fall. It changes composition. Access and cost tickets grow. Pipeline-fix tickets shrink. The net is roughly flat, and the skill mix shifts toward platform primitives.

Second, domain teams underestimate the coordination cost of federated governance. Every cross-domain schema change requires a conversation, and every conversation has a latency. The latency is the tax.

Third, the tickets that hurt most are the ones with no clear owner: CDC gaps, expired snapshots, and schema breaks that cross a domain boundary. These are the tickets that sit in the queue the longest, because no single team’s on-call rotation owns them.

What to do with the count

The count is not an argument against data mesh. It is an argument for naming the ownership boundary explicitly, in writing, for the three failure modes that cross domains: CDC gaps, snapshot expiry, and cross-domain schema changes.

For each, write down: who detects it, who decides the recovery, who executes the backfill, and who pays for the compute. If any of those four is “the platform team” by default, the mesh is not fully implemented. It is a mesh with a central fallback, and the fallback is the queue.

The platform team’s ticket queue is the honest scoreboard. Read it before you read the architecture diagram.

FAQ

How do I tell new work from reclassified work?
Compare platform-team hours by activity, not ticket count. If you did not capture a pre-rollout baseline, you cannot make the distinction. Capture it now and compare forward.

Does data mesh reduce platform load?
The principles do not promise that. They promise to move ownership of analytical data to domain teams. Platform load shifts from pipeline fixes to platform primitives. The net depends on how many domains onboard and how much of the control plane is centralized.

What is the cheapest guardrail against schema breaks?
dbt data tests. A not_null test on a column that should never be null, and a relationships test on a foreign key, will catch most renames and drops before consumers see them. Both are SQL queries that return failing rows.

Who owns a CDC gap?
Nobody, by default. That is the problem. Assign detection, decision, execution, and cost explicitly, or the ticket will sit in the platform queue.

Is clustering a domain decision or a platform decision?
It is a domain decision with a platform cost. Snowflake’s documentation is clear that clustering is not for all tables and that reclustering consumes credits and storage. Treat the clustering key as a budget decision, not a performance decision.

Sources

• Zhamak Dehghani, “Data Mesh Principles and Logical Architecture,” martinfowler.com, 03 December 2020. https://martinfowler.com/articles/data-mesh-principles.html
• Data Mesh Architecture, “What Is Data Mesh?” https://www.datamesh-architecture.com/
• dbt Docs, “Data tests.” https://docs.getdbt.com/docs/build/data-tests
• Apache Airflow Documentation, “Dynamic Task Mapping.” https://airflow.apache.org/docs/apache-airflow/stable/authoring-and-scheduling/dynamic-task-mapping.html
• Apache Iceberg Table Spec. https://iceberg.apache.org/spec/
• Snowflake Documentation, “Clustering Keys & Clustered Tables.” https://docs.snowflake.com/en/user-guide/tables-clustering-keys