Hello community, I reported an issue regarding the Connector Builder / Declarative CDK a while ago, and I would heavily appreciate it if someone from the team could take a look, as it severely impacts sync performance. When using

incremental_dependency: true

on parent-child streams, the child stream drops its cursor context after the first sync, forcing the parent stream into an unexpected full refresh loop on subsequent syncs (making them 100x–1000x slower). I just updated the issue with a comprehensive root-cause analysis (assisted by GitHub Copilot and verified against the latest Concurrent CDK source code). I also provided: 1. A functional connector-level workaround using custom components (StateMigration) that anyone can apply today. 2. The exact lines in the CDK core (

substream_partition_router.py

) where the context is being lost. Here is the link to the issue: github.com/airbytehq/airbyte/issues/61567 Could you please help review the proposed architectural fix or provide recommendations to get this merged? Thanks!

<!subteam^S01SHT8BSCU> are we wasting our times trying to maintain community connectors? Should we build something decentralised where people can just copy-paste the latest connector instead?

Is airbyte planning on supporting connection directly to Microsoft Fabric Warehouse ??? Has anyone been able to connect it?

Hi team 👋 . Could I get a maintainer review on a source-gong fix? PR #82260 — github.com/airbytehq/airbyte/pull/82260 Fixes #78480 (source-gong 429 / rate-limit failures) What it does:

callTranscripts

currently sends one POST /calls/transcript request per call (SubstreamPartitionRouter with a single callId), so backfills blow past Gong’s ~3 req/s rate limit and ~10k/day quota. The PR wraps the existing router in a GroupingPartitionRouter (group_size: 100) so one request covers up to 100 calls — ~99% fewer requests. Manifest-only change + a unit test; preserves the private-call filter and started cursor, no schema change/reset. Validated against a live Gong tenant: array callIds accepted, 100/request → 200, cursor pagination works alongside callIds. Status: CLA signed ✅, all fork-accessible checks green. The only red checks are the fork “No Creds” connector test and the progressive-rollout gate — both need maintainer-side action (/run-connector-tests + rollout bypass). Would appreciate someone running the credentialed suite and taking a look 🙏 (edited)

hey i made a pr to add jwt auth for salesforce, can someone help me get this reviewed? github.com/airbytehq/airbyte/pull/80788

Hi team, I submitted an issue a month ago, and the AI autoreply said they would have the issue routed to oncall, but I have no clue how to track the oncall progress. This one is fairly important for us github.com/airbytehq/airbyte/issues/80250 any ideas?

Hi team! 👋 I've been investigating a silent data-loss bug in

source-snowflake

(currently on

1.0.5

, but I confirmed it's still reproducible on

1.1.0

) reported here: github.com/airbytehq/airbyte/issues/70257 (filed by someone else back in December). Incremental syncs return 0 records when the stored cursor checkpoint equals the current

MAX(cursor_field)

, with no error. Root cause looks like:

SnowflakeLocalDateTimeAccessor

in

SnowflakeFieldTypes.kt

truncates nanosecond precision down to microseconds when reading the

MAX(cursor)

value used to build the query's upper bound, but the actual column values in Snowflake keep full nanosecond precision. When the sync window collapses to a single point (

cursor == MAX

), the truncated upper bound ends up strictly less than the real value of the row(s) it's meant to include, so the WHERE clause silently excludes them. The fix would live entirely in the connector's own source, not the shared Bulk CDK. Before I put together a PR: I noticed the contributing guide mentions community contributions to "the data replication platform" stopped being accepted in mid-2025, and the accepted-contributions list is framed around API/Connector-Builder connectors. Since Snowflake is a JDBC/Kotlin connector, is a bug fix like this still in scope for a community PR, or should this go through another channel? Happy to open the PR either way if someone can point me in the right direction. Thanks!

Hi! I opened PR #82690 for issue #27171, adding an optional JSONL stringify mode for Hive JsonSerDe compatibility. The Bulk CDK checks pass. The destination compatibility jobs fail before running because the fork PR lacks connector test configuration, and I asked for confirmation in the PR. Could a maintainer confirm whether this CDK/destination-related contribution is in scope for review and whether the compatibility jobs can be rerun with the required configuration?

hello 👋 the source-postgres connector suffers from lots of bugs since

3.8.x

2 months old release, examples: • github.com/airbytehq/airbyte/issues/80939 • github.com/airbytehq/airbyte/issues/80295 • github.com/airbytehq/airbyte/issues/81431 • github.com/airbytehq/airbyte/issues/81397 • github.com/airbytehq/airbyte/issues/47547 • github.com/airbytehq/airbyte/issues/79109 👉 do you have any visibility regarding a solid fix ? ⚠️ this is blocking for this 🏎️ speeded up destination v2 adoption as

Change Data Capture

is unusuable 😕 thank you!

Hey everyone! Does anyone experience problem with Paypal and Bing connectors? We started getting this error all of a sudden. We use native connectors.

Hey everyone, I am trying to create a custom connector for Reddit Ads API (Endpoint: ads-api.reddit.com/docs/v3/api/get-a-report) using Connector Builder UI. Has anyone done it before?

Hello! A new column called _network_details_ was added to the Stripe's disputes table, we tried hitting their API end point and it is indeed available. However, after refreshing our connector's source schema in Airbyte this column did not appear. We also tried updating our Stripe connector which also did not solve the issue. Perhaps anyone knows how we could add this new column to our Stripe connector and how long this process might take?

Hi everyone, I'm researching ways to use AI agents to improve the connector development experience. I'm wondering: - Does Airbyte have an open-source implementation for AI-assisted connector generation? - Which repository contains the related code? - Is it based on Airbyte Agent SDK, Connector Builder, or other internal components? I'd like to contribute ideas and improvements around: - API documentation understanding - schema inference - connector YAML/config generation - developer experience improvements If there is a contributor group or discussion channel for this topic, I'd appreciate any guidance. Thank you!

Hi Team, I submitted a PR for updates to the Uptick source connector: github.com/airbytehq/airbyte/pull/83710 Could I please request a review

Anybody uses MySQL as destination? There are no docs about why 0 precision columns from Snowflake create decimal(38,9( rather than integer?

Hi Team! I have a PR open on source-float since June that's still waiting for a review: github.com/airbytehq/airbyte/pull/79087 It adds two streams from the Float Reports API —

reports-people

and

reports-projects

(aggregated scheduling/utilization data) — plus an optional

end_date

config field. Manifest-only, 0.0.56 → 0.1.0, existing streams untouched. I've just merged latest master, so it's conflict-free and mergeable again. Would appreciate a review whenever someone has time — happy to make changes.

Hi team 👋 We have several open PR waiting for Airbyte team approval, please could you consider them ? 🙏 🔴 Blocking data collect (current connector is down) • Goldcast new API interface 🟠 Blocking our upgrade to speed-mode GCS destination v1.0.0 (fields regressions) • Goldcast event_members schemaless • Hubspot marketing_email &amp; owners missing fields 🟡 Blocking new business needs implementations • LinkedIn Ads video stream • Chift executions stream and schemaless management Thanks a lot, and happy to help 🙂

Hi Team/@Richelle Herrli I've been facing an issue where the data movement from GCS to Snowflake creates junk column on every run when authenticated with Service Account. The issue was raised a couple of months back here. Since there was no resolution provided, I made the changes and raised a PR, can somebody please help in reviewing this so that we can be unblocked? Thank you!

Hi team 👋, any plan for this issue github.com/airbytehq/airbyte/issues/82729? I think the AI integration code is getting there github.com/airbytehq/airbyte/pull/82731

Hi everyone, I’ve submitted a feature request to Airbyte for configuring CPU and memory resources per connection directly from the UI. Airbyte already supports connection-specific resource requirements internally, but today this typically requires changing the internal database manually. In environments with many connections, this makes resource management harder than it should be. The main use case is that not every connection has the same workload. For example, a lightweight CDC connection may only need a small amount of CPU and memory, while another connection doing a large initial sync or heavy batch load may need much more. Being able to configure these values per connection from the UI would help avoid both over-provisioning and under-provisioning. If this would be useful in your environment as well, please consider upvoting the feature request. More upvotes can help show the Airbyte team that there is broader demand for this capability. github.com/airbytehq/airbyte/discussions/85061

Hey everyone 👋 We’re running Airbyte 2.0 and during routine container scanning we’ve identified several OS-level vulnerabilities in Airbyte-provided base images: •

airbyte/container-orchestrator

•

airbyte/connector-sidecar

•

airbyte/workload-init-container

The vulnerabilities appear to originate from the Java base image bundled with these containers — so they’re outside our control to patch directly. We’ve followed the reporting process outlined in the Airbyte security docs and sent the details to

<mailto:security@airbyte.io|security@airbyte.io>

, and even after multiple follow ups, haven’t received a response yet. A couple of questions for the community: 1. Has anyone else run into OS-level CVEs in these Airbyte core images? If so, how did you get them addressed? 2. Is there a better channel to reach the Airbyte security team — GitHub issue, Slack, or otherwise? Any pointers would be really appreciated! 🙏

Hi all, looking for a maintainer to unblock a community connector PR. PR: airbytehq/airbyte#82708 (

feat(source-cursor): 🎉 New Source: Cursor

), open since 23 July. Manifest-only source for the Cursor Teams Admin API. Everything passes except two checks:

source-cursor Progressive Rollout Gate

and

Connector Active Progressive Rollout Checks Summary

. Both fail with:

Error: GCP_PROD_DB_ACCESS_CREDENTIALS environment variable is not set.

That's an Airbyte-internal secret, so it can't run on a fork PR. This is also a brand new connector with no active rollout, so there shouldn't be anything for the gate to check. I left a comment on the PR on 27 August but haven't heard back. Could someone re-run those with access, or waive the gate if that's the right call? Happy to make any other changes needed.

Hi all I've opened a new StarRocks destination connector for the monorepo and would appreciate a maintainer's eyes on it. • Issue: github.com/airbytehq/airbyte/issues/86474 (follows up on discussion #53631) • PR: github.com/airbytehq/airbyte/pull/86475 It's written in Kotlin on the Bulk CDK (core = load, Direct Load, cdkVersion=1.1.0) Three things I could use help with: 1. definitionId –

metadata.yaml

currently carries the UUID from the StarRocks/airbyte-starrocks repo (5c4d966a-19ff-45d8-9687-876ad0f5d0d9). Is a contributor-set UUID fine, or does someone on the Airbyte side need to issue or register one? 2. Policy – the contributing guide still says "no new Java connectors". Does that apply to Kotlin / Bulk CDK destinations? (Same question chris-celerdata asked in #53631 and never got an answer.) Thanks!

Hi everyone, When an Airbyte sync fails prematurely (e.g., due to an image pull failure from the registry) with 0 records synced, Airbyte is still reporting the status as Successful. This false-positive status makes it easy to miss silent failures. I noticed an open issue tracking this on GitHub (Replication pod creation fails (KubeClientException) but UI/Public API still reports sync as successful · Issue #72448 · airbytehq/airbyte), but in the meantime, I wanted to check: how are others currently detecting and handling these false-positive syncs? Any workarounds or monitoring approaches you’d recommend?

Hi Airbyte team 👋 To fully benefit from

speed mode (sockets)

on below source connectors using

Change Data Capture

, we are still facing some blocking issues. Please could you consider fixing them ASAP ? 🙏 PostgreSQL • github.com/airbytehq/airbyte/issues/80939 ⏳ • github.com/airbytehq/airbyte/pull/86369 ✅ • github.com/airbytehq/airbyte/pull/86976 ✅ • github.com/airbytehq/airbyte/pull/86921 ⏳ • github.com/airbytehq/airbyte/issues/82266 ⏳ • github.com/airbytehq/airbyte/pull/84996 ⏳ • (and github.com/airbytehq/airbyte/issues/81431 for temporary XMin migrations management to CDC) ⏳ MSSQL • github.com/airbytehq/airbyte/pull/85353 ⏳ • github.com/airbytehq/airbyte/pull/86413 ⏳ • github.com/airbytehq/airbyte/pull/86976 ⏳ Thank you a lot ! 🔥 FYI @Richelle Herrli @Justin Beasley @Ian Alton 💯

Hello Team, we were testing clickhouse version 26.8lts for the improvements in the performance and other various features, it seems the airbyte clickhouse connector is not compatible with the clickhouse version 26.8 , it seems some issues in the bulk module where the table structure is not fetched , as the column name and column values for the table are null. is someone also facing issues with the new version? any feedback or direction would be really helpful.

Hi, I opened a new manifest-only source for the OpenAI Ads API three weeks ago github.com/airbytehq/airbyte/pull/86351. CI is green apart from the fork-gated rollout checks and it's still in New PRs on the community board. Is there anything I can do to get it into a review sprint?

Hi! I opened a PR adding a

Listings

stream to source-hubspot (HubSpot's Listings CRM object): github.com/airbytehq/airbyte/pull/88970 (resolves #88969). CLA signed, CI is green, and it's in New PRs on the community board. I also ran the full connector test suite on my own fork with BYO credentials as recommended. Could a maintainer run

/run-connector-tests

when they get a chance? I'd love to get this reviewed soon!

Subject: destination-bigquery fails the whole attempt on a transient BigQuery 503 while polling a load job Hi team. We run Airbyte OSS (destination-bigquery 3.0.7, GCS Staging) syncing a MySQL source (full refresh, ~131M records / 43 GB per day) into BigQuery. On two consecutive days the sync took 2-4 hours longer than usual. The cause was not the sync speed: an attempt failed while the destination was waiting for a GCS -> BigQuery load job, and Airbyte then re-ran the attempt from scratch. The failure comes from BigQueryUtils.waitForJobFinish -> job.waitFor(...): the jobs.get polling call got an HTTP 503 (BigQueryError{reason=backendError, message=Error encountered during execution. Retrying may solve the problem.}) while the load job itself was still RUNNING. The connector raises RuntimeException("Failed to load CSV data from gs://..."), the attempt is marked failed, and the next attempt restarts the full sync (the attempt that eventually succeeded took the normal ~4 hours, so nothing from the failed attempts was reused). Example (job 4058, 2026-10-20 UTC): - attempt 0: failed after 56 min (10 GB of 43 GB emitted) - 503 on jobs.get for the hearing_other_routine_brands load - attempt 1: failed after 78 min - 503 on jobs.get for the pos_purchases load - attempt 2: succeeded in 3 h 57 min Questions: 1. Is it intended that a transient 5xx on jobs.get fails the attempt instead of re-polling the same (still running) load job? Google's docs say a backendError on jobs.get should simply be retried, and the job can be re-polled without data-consistency concerns. 2. I found github.com/airbytehq/airbyte/pull/87606 (retry transient BigQuery 5xx via the client's retry algorithm), which looks like the same signature. Is there a plan to merge/release it, and would it cover the job.waitFor polling path in the GCS Staging bulk loader? 3. Until then, is there any configuration (connector or platform) that makes the destination retry the load-job polling rather than restarting the attempt? We are currently considering a fork of 3.0.7 that re-polls the same job with exponential backoff on 5xx/backendError. Happy to share full logs. Thanks!

Hi all, I filed a source-mssql CDC bug and would like a maintainer's read on the fix before I open a PR: https://github.com/airbytehq/airbyte/issues/90141 Short version: if the newest transaction at sync start has many changes, the CDC read stops at its first change (

RECORD_REACHED_TARGET_POSITION

, since

MsSqlServerCdcPosition

compares only

commit_lsn

). The rest isn't read until the max LSN moves. So a transaction that moves rows between two tables lands half-applied in the destination. It reproduces on 5.1.2 (54 of 5,000 rows arrive on the first sync). It also reproduces on the #88991 pre-release with CDK 1.1.22, which delivers more (1,664 of 5,000) but still splits. The issue has the repro and two candidate fixes. I'm happy to implement whichever you prefer.