Product1 distinct publisher3 min readUpdated
Yugabyte says Sequin, PeerDB and Supabase ETL each needed only minor changes, and nearly all of them trace to log positions that are not global coordinates.
The Product Desk · Product desk

Compiled by The Product DeskSomething wrong?How this is made
Yugabyte pointed three popular open-source PostgreSQL change data capture tools, Sequin, PeerDB and Supabase ETL, at YugabyteDB and reported that all three worked, each after a small and easy-to-execute set of changes [1][2]. The interesting part for anyone outgrowing a single Postgres box is where the work landed: not in migrating the database, but in a short audit of assumptions inside the connector [9].
The reason the tools run at all is that YugabyteDB uses PostgreSQL's own logical replication, with the familiar replication slots, publications and per-change detail settings [4]. According to Yugabyte, the everyday live streaming path came up for all three tools with no changes to the tools themselves [5]. Yugabyte says the whole exercise took a few weeks, most of it spent reading code and writing tests rather than fighting the database [3].
The one concept that generates the fixes is the Virtual WAL. A single-node PostgreSQL server keeps one write-ahead log, so any two positions in it can be lined up and compared, and plenty of tools quietly assume exactly that [6]. YugabyteDB spreads a table across many pieces, each with its own log, then merges them back into commit order and hands the connector something that looks like an ordinary log [7]. The catch is that a position in one YugabyteDB stream only means something inside that stream; Yugabyte's logical replication documentation states these are not global coordinates and cannot be compared across streams [8].
Almost every change a tool needed comes from that [9]. A few PostgreSQL helper functions assume the single global log and do not carry over, and any logic that compares positions across two separate streams has to be rethought [9]. Where a tool genuinely needs a marker comparable across the whole database, YugabyteDB provides one [10]. A couple of further differences follow from YugabyteDB storing data in its own distributed layout rather than the classic single-machine one; per Yugabyte, those touched only one of the three tools and the fixes were minor [11]. Yugabyte points readers at its key concepts and limitations pages before starting [15].
Sequin, an open-source CDC platform that streams Postgres changes to destinations such as Kafka [12], is the case broken out in detail: with standard setup, both live streaming and the initial backfill came up without changes to Sequin's core logic, and the changes it did require applied to the Virtual WAL and to one or two older PostgreSQL features YugabyteDB does not support [13][14].
Two things to hold in mind. This is the database vendor testing compatibility with its own product, and the material available stops partway through the Sequin section, before comparable per-tool detail for PeerDB or Supabase ETL [17]. And "worked with small changes" is not the same as "worked": each of the three needed patches, even though the streaming path itself needed none [16]. Anyone costing a migration should price the fork, the ongoing rebase against upstream, and the test suite that proves position handling behaves.
What to watch: whether the per-tool change sets are published in full and land upstream in Sequin, PeerDB and Supabase ETL rather than living in Yugabyte's own branches [14][17], and whether the older unsupported PostgreSQL features named in the Sequin work turn out to matter to other connectors [14].
Follow any of these and your For You feed starts watching them — no settings page required.
Ranked by verification strength, evidence, and original report placement.
All three tools worked with YugabyteDB; each needed a small, easy-to-execute set of changes, and every one of those changes came back to the same idea.
The changes Sequin required applied to the Virtual WAL and to one or two older PostgreSQL features that YugabyteDB does not support; Yugabyte describes them as small.
The blog states it will show how each tool performed with YugabyteDB, but the text available ends partway through the Sequin section, before equivalent per-tool detail for PeerDB and Supabase ETL and before the complete list of Sequin's required changes.
Yugabyte took three popular open-source PostgreSQL CDC tools - Sequin, PeerDB and Supabase ETL - connected them to YugabyteDB, ran them under real-life conditions, and published the findings.
Yugabyte tested all three CDC tools within a few weeks, and most of that time was dedicated to reading code and writing tests rather than wrestling with the database.
YugabyteDB uses PostgreSQL's own logical replication, including replication slots that track where a consumer is in the stream, publications that specify which tables and changes matter, and the usual settings for how much detail each change carries.
Evidence-backed comparisons of source perspectives and observed adoption signals. Read the methodology
Which Builder, Operator, and Investor concerns the observed source mix emphasized—not a truth score.
Evidence, demonstrated adoption, hype gap, incentives, and confidence are assessed independently, each on its own current evidence. How these are measured.
Detailed but single-source and self-assessed
The cluster contains one primary source: the vendor's own engineering blog. It is unusually specific for vendor material - it names the mechanism (Virtual WAL, stream-local positions), the exact class of code changes, a linked public fork, and before/after test-suite failure counts - which lifts it above pure assertion. But there is no independent corroboration from Sequin, PeerDB or Supabase, no reproducible harness or configuration published in the text, and the body truncates before the Supabase ETL result, so the three-tool generalisation is only partly evidenced.
Vendor-run experiments and one fork
Observable adoption is limited to Yugabyte's own lab work: a fork it maintains, a test-suite run, and a PeerDB-to-Redpanda pipeline it built itself. No third-party user, customer, production workload or upstream merge is reported, and no Supabase ETL outcome is published at all, so real-world uptake of PostgreSQL CDC tools on YugabyteDB remains unevidenced.
Headline outruns published results
The framing that all three tools worked with only small changes is broader than what the available text demonstrates: Supabase ETL results are absent, the 14 acknowledged feature gaps are not enumerated, 'small' is the vendor's own characterisation, and the unmodified-out-of-the-box result applies only to the live streaming path while each tool still needed patches. The gap is moderate rather than severe because the piece is candid about the Virtual WAL constraint, publishes failure counts that include remaining failures, and links a fork readers can inspect.
Vendor testing its own compatibility story
The sole source is Yugabyte's marketing-adjacent engineering blog, assessing whether third-party PostgreSQL tools work with the product Yugabyte sells. PostgreSQL compatibility is the company's central differentiator, so a favourable outcome directly serves its commercial interest, and Yugabyte controls both the test design and which results are published. Partially mitigating: the post concedes unsupported features, publishes failure counts including residual failures, and links a fork for inspection.
Plausible mechanism, thin verification
The technical explanation is internally coherent and matches how distributed logical replication has to work, and the fork plus test numbers give something concrete to check. Confidence is nonetheless capped by a single self-interested publisher, no independent replication, no operational metrics, and a truncated body that omits one of the three tool results.
build
Three services you can delete: queue, cache and search in one Postgres1 distinct publisher
build
Notion's agent stack is live, not slideware, and it only changes one of your decisions1 distinct publisher
build
The 680 MB database that was really a 17 GB disk: self-hosted support platforms fail at month six1 distinct publisher
build
Your "Index Only Scan" Did 2,847 Heap Fetches: Covering Indexes Are a Vacuum Problem1 distinct publisher
Distinct publishers with included, body-backed reporting in this cluster.
1 article · August 17, 2026