Deprecate non-ClickHouse / non-Postgres destination connectors

[Jeremyyang920]

Implements the code/config deliverables for #4465 — signaling the deprecation of the non-ClickHouse / non-Postgres destination connectors. No connector code is removed; deprecated connectors stay fully functional.

Deprecated destinations: Snowflake, BigQuery, ElasticSearch, Kafka (incl. Confluent + Redpanda), Azure Event Hubs, Google Pub/Sub, and S3. Actively-maintained paths: Postgres → ClickHouse / ClickHouse Cloud / Postgres.

Changes

• README — deprecation notice placed just after the intro (above the fold) + a connector status matrix. BigQuery is shown as a maintained source and deprecated destination.
• docs/deprecated-connectors.md (new) — pin + fork migration guide, per-connector paths.
• CONTRIBUTING.md (new) — links to the migration guide.
• UI (peer-types route + SelectSource) — uniform "Deprecated" badge in the peer-creation grid. Since peers are direction-agnostic and BigQuery is a valid source, BigQuery is marked "Deprecated*" with a footnote: "Deprecated as a destination only."
• GitHub automation — issue form with a destination-connector dropdown + deprecation notice; .github/labeler.yml + deprecated-connector.yml workflow that labels deprecated-connector PRs and posts a single de-duplicated notice (never auto-closes).

Notes / follow-ups (not in this PR)

• The deprecated label must be created once in repo Settings → Labels (labels aren't defined in-repo).
• Post-merge: apply the deprecated label + standard comment to existing open non-CH destination issues/PRs (do not auto-close).
• BigQuery is deprecated only as a destination — it remains a supported source.

Verification

• Brought the full dev stack up (after generate-protos.sh); verified /api/peer-types returns the correct per-connector deprecated/deprecatedRole data and /peers/create renders the badges + footnote without errors.
• All issue-form / labeler / workflow YAML validated.
• GitHub automation validated end-to-end on a fork (see the pinned summary comment).

🤖 Generated with Claude Code

[github-actions]

🔄 Flaky Test Detected

Analysis: The deterministic unit test TestRunPipeline_FilterStripsLines failed with "src failed: signal: killed" — its trivial printf subprocess was SIGKILLed (OOM/resource pressure) under the heavily parallel CI run, not due to a code bug.
Confidence: 0.88

✅ Automatically retrying the workflow

View workflow run

[Jeremyyang920]

Tested the GitHub automation end-to-end on a fork (Actions enabled, branch as default, deprecated label created). This surfaced a real bug in the labeler config from my previous commit:

The all: + all-globs-to-all-files negation block silently disabled the entire actions/labeler@v5 config — the label job succeeded but applied no labels at all (even a trivial flow/connectors/snowflake/** match was skipped). No error is logged; the action just no-ops.

Fix (d78c1d7b): dropped the all: block and instead list BigQuery's destination-specific files explicitly, so BigQuery source changes (source.go, source_schema.go, qrep_object_pull.go, bigquery_source_test.go) are still never auto-labeled.

Verified on the fork after the fix:

Scenario	Result
PR touching flow/connectors/snowflake/**	✅ deprecated label + notice comment
PR touching only BigQuery source files	✅ not labeled, no comment
PR touching BigQuery destination (bigquery.go)	✅ deprecated label + notice comment
Manual deprecated label on an issue	✅ notice comment posted

The needs: label + if: always() change on the notice job also confirmed working: on pull_request_target the notice posts after the labeler applies the label.

[codecov]

❌ 2 Tests Failed:

Tests completed	Failed	Passed	Skipped
2919	2	2917	338

View the top 2 failed test(s) by shortest run time

github.com/PeerDB-io/peerdb/flow/e2e::TestPeerFlowE2ETestSuiteMySQL_CH_Cluster

Stack Traces | 0.01s run time

=== RUN   TestPeerFlowE2ETestSuiteMySQL_CH_Cluster
=== PAUSE TestPeerFlowE2ETestSuiteMySQL_CH_Cluster
=== CONT  TestPeerFlowE2ETestSuiteMySQL_CH_Cluster
--- FAIL: TestPeerFlowE2ETestSuiteMySQL_CH_Cluster (0.01s)

github.com/PeerDB-io/peerdb/flow/e2e::TestPeerFlowE2ETestSuiteMySQL_CH_Cluster/Test_MySQL_Default_Partition_Key_Parallel_Snapshot

Stack Traces | 5.23s run time

=== RUN   TestPeerFlowE2ETestSuiteMySQL_CH_Cluster/Test_MySQL_Default_Partition_Key_Parallel_Snapshot
=== PAUSE TestPeerFlowE2ETestSuiteMySQL_CH_Cluster/Test_MySQL_Default_Partition_Key_Parallel_Snapshot
=== CONT  TestPeerFlowE2ETestSuiteMySQL_CH_Cluster/Test_MySQL_Default_Partition_Key_Parallel_Snapshot
2026/07/06 15:22:08 INFO Received AWS credentials from peer for connector: ci x-peerdb-additional-metadata={Operation:FLOW_OPERATION_UNKNOWN}
2026/07/06 15:22:08 INFO Received AWS credentials from peer for connector: clickhouse x-peerdb-additional-metadata={Operation:FLOW_OPERATION_UNKNOWN}
2026/07/06 15:22:08 INFO fetched schema x-peerdb-additional-metadata={Operation:FLOW_OPERATION_UNKNOWN} table=e2e_test_mychcl_zse7migc.test_mysql_ghost_schema
    clickhouse_mysql_test.go:1971: WaitFor waiting on initial 2026-07-06 15:22:13.739667231 +0000 UTC m=+871.966058476
2026/07/06 15:22:13 INFO fetched schema x-peerdb-additional-metadata={Operation:FLOW_OPERATION_UNKNOWN} table=e2e_test_mychcl_xbxbydfv.test_default_partition_key_parallel
    clickhouse_mysql_test.go:1983: 
        	Error Trace:	.../flow/e2e/clickhouse_mysql_test.go:1983
        	Error:      	Received unexpected error:
        	            	can't scan into dest[0] (col: rows_in_partition): cannot scan NULL into *int64
        	Test:       	TestPeerFlowE2ETestSuiteMySQL_CH_Cluster/Test_MySQL_Default_Partition_Key_Parallel_Snapshot
--- FAIL: TestPeerFlowE2ETestSuiteMySQL_CH_Cluster/Test_MySQL_Default_Partition_Key_Parallel_Snapshot (5.23s)

To view more test analytics, go to the Test Analytics Dashboard
_{📋 Got 3 mins? Take this short survey to help us improve Test Analytics.}

[github-actions]

🔄 Flaky Test Detected

Analysis: The lone failure, TestRunPipeline_FilterStripsLines, died with "src failed: signal: killed" (empty stderr) on a trivial in-memory printf→cat pipeline that touches no DB/network — a classic OOM/resource-exhaustion kill under -p 32 high concurrency, unrelated to the PR's destination-connector deprecation change.
Confidence: 0.92

✅ Automatically retrying the workflow

View workflow run

[github-actions]

🔄 Flaky Test Detected

Analysis: Test_MySQL_BinlogIncident failed on a transient MySQL handshake EOF ("connection was bad") from its ephemeral debug testcontainer — a network/connection flake unrelated to this destination-connector-deprecation PR.
Confidence: 0.85

✅ Automatically retrying the workflow

View workflow run

[Jeremyyang920]

✅ Current scope + validation (supersedes earlier notes)

Posting as a comment rather than editing the PR description, per this repo's CLAUDE.md policy ("never modify PR descriptions/titles; report changes as PR comments"). The description's note that "S3 is intentionally not deprecated" is now out of date — see below.

Deprecated destination connectors (final scope)

Snowflake, BigQuery, ElasticSearch, Kafka (incl. Confluent/Redpanda), Azure Event Hubs, Google Pub/Sub, and S3.

• S3 is now deprecated (destination-only; it isn't a source). README matrix, docs, the UI badge, the labeler, and the issue-form dropdown/notice all include it.
• BigQuery is deprecated only as a destination — it remains a supported source (badge reads "Deprecated*" with a footnote; labeler excludes BigQuery source files).
• ClickHouse / ClickHouse Cloud / Postgres (and Postgres/MySQL/MongoDB sources) remain actively maintained.

Validated end-to-end on a fork (Jeremyyang920/peerdb)

Actions enabled, branch set as default, deprecated label created.

Path-based PR labeling (actions/labeler)

• PR #2 — flow/connectors/snowflake/** → labeled + notice ✅
• PR #3 — only BigQuery source (source.go) → not labeled ✅
• PR #4 — BigQuery destination (bigquery.go) → labeled + notice ✅

Issue-form dropdown labeling

• Issue #5 — selects Snowflake → labeled + notice ✅
• Issue #6 — selects ClickHouse (maintained) → not labeled ✅
• Issue #7 — selects BigQuery (destination) → labeled + notice ✅

Manual labeling

• Issue #1 — maintainer adds deprecated → notice posted ✅

Notice comments are de-duplicated (hidden marker) and nothing is ever auto-closed. (S3 follows the same destination-only mechanism already proven for Snowflake/ElasticSearch above.)

One-time setup before this works upstream

Create the deprecated label in PeerDB-io/peerdb → Settings → Labels.

[Jeremyyang920]

Thanks @ilidemi — addressed in b8fbdbe:

• labeler.yml header — now lists S3 as deprecated and notes ClickHouse/Postgres as the excluded (actively-maintained) destinations.
• labeler.yml (line ~47) — applied your suggestion: added flow/connectors/bigquery/qrep.go to the tracked BigQuery destination files.
• CONTRIBUTING.md — added S3 to the deprecated list (and noted BigQuery is deprecated only as a destination).
• bug_report.yml — removed the duplicate migration-guide link.
• config.yml / feature requests — good catch; added a feature_request.yml template so feature requests have a proper home while keeping the chooser curated (blank issues stay off).
• README → linked status doc — that's the external docs.peerdb.io supported-connectors page; it'll be updated separately as part of this effort, so no in-repo change here.

[claude]

Code review

No issues found. Checked for bugs and CLAUDE.md compliance.

[github-actions]

🔄 Flaky Test Detected

Analysis: The lone failing test TestRunPipeline_FilterStripsLines died with "src failed: signal: killed" — its trivial printf subprocess was externally SIGKILLed (OOM/resource contention on the heavily-loaded -p 32 runner), which is an environmental flake rather than a logic bug.
Confidence: 0.9

✅ Automatically retrying the workflow

View workflow run

[github-actions]

🔄 Flaky Test Detected

Analysis: The test reads qrep_partitions.rows_in_partition right after data lands in ClickHouse, but that column is populated by a pull goroutine that races the concurrent sync goroutine, so it can still be NULL — confirmed intermittent since the identical query passed in sibling tests and in the same test on other matrix jobs.
Confidence: 0.85

✅ Automatically retrying the workflow

View workflow run