feat(pg-pg): Automated schema dump mode

[Amogh-Bharadwaj]

Why

• For Postgres to Postgres migration use-cases, users would prefer an apples-to-apples mirroring of their source setup on the target instance.
• A well-recognized method to migrate schema only is pg_dump.

What

• This PR adds an activity in SetupFlowWorkflow which runs pg_dump on the source database and pipes its output to the target database via PSQL.
• It adds a new dynamic flag setting to gate the above. Also the above is gated to PG type system mirrors.
• When the above flag is set, destination validation and destination create normalized tables steps are skipped.
• It sets --no-owners and --no-privileges, leaving it to the user to add the desired roles later. pg_dumpall requires exact major version matching between pg_dumpall version and target version, which we cannot guarantee.
• E2E tests added

[codecov]

❌ 2 Tests Failed:

Tests completed	Failed	Passed	Skipped
2243	2	2241	203

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

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

Stack Traces | 0.01s run time

=== RUN   TestApiPg
=== PAUSE TestApiPg
=== CONT  TestApiPg
--- FAIL: TestApiPg (0.01s)
2026/05/13 18:52:34 INFO Received AWS credentials from peer for connector: ci x-peerdb-additional-metadata={Operation:FLOW_OPERATION_UNKNOWN}
2026/05/13 18:52:34 INFO Received AWS credentials from peer for connector: clickhouse x-peerdb-additional-metadata={Operation:FLOW_OPERATION_UNKNOWN}

github.com/PeerDB-io/peerdb/flow/e2e::TestApiPg/TestResyncCompleted

Stack Traces | 0.09s run time

=== RUN   TestApiPg/TestResyncCompleted
=== PAUSE TestApiPg/TestResyncCompleted
=== CONT  TestApiPg/TestResyncCompleted
2026/05/13 18:50:49 INFO Received AWS credentials from peer for connector: ci x-peerdb-additional-metadata={Operation:FLOW_OPERATION_UNKNOWN}
2026/05/13 18:50:49 INFO Received AWS credentials from peer for connector: clickhouse x-peerdb-additional-metadata={Operation:FLOW_OPERATION_UNKNOWN}
    api_test.go:1302: 
        	Error Trace:	.../flow/e2e/api_test.go:1302
        	Error:      	Received unexpected error:
        	            	failed to get workflow_id for flow resync_completed_api_3hfhot8l: FATAL: terminating connection due to administrator command (SQLSTATE 57P01)
        	Test:       	TestApiPg/TestResyncCompleted
    api_test.go:49: begin tearing down postgres schema api_3hfhot8l
--- FAIL: TestApiPg/TestResyncCompleted (0.09s)

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]

❌ Test Failure

Analysis: Consistent, deterministic failures across all CI matrix jobs caused by pg_dumpall: error: aborting because of server version mismatch in Test_PG_Schema_Dump_Role_Migration, with cascading ~31s timeouts in multiple other TestPeerFlowE2ETestSuitePG tests — a real environment/version mismatch bug, not flaky.
Confidence: 0.93

⚠️ This appears to be a real bug - manual intervention needed

View workflow run

[github-actions]

🔄 Flaky Test Detected

Analysis: Multiple e2e tests in TestPeerFlowE2ETestSuitePG failed with "UNEXPECTED STATUS TIMEOUT STATUS_SETUP" at ~31 seconds across all CI matrix jobs, indicating the test infrastructure was unable to complete setup in time rather than a code regression — consistent with CI resource contention when running 32 parallel tests.
Confidence: 0.92

✅ Automatically retrying the workflow

View workflow run

[github-actions]

❌ Test Failure

Analysis: Multiple TestPeerFlowE2ETestSuitePG tests consistently fail with UNEXPECTED STATUS TIMEOUT STATUS_SETUP across all three CI matrix variants, suggesting the recent docker-compose stable image tag upgrade broke flow setup rather than a random flaky timeout.
Confidence: 0.78

⚠️ This appears to be a real bug - manual intervention needed

View workflow run

[github-actions]

🔄 Flaky Test Detected

Analysis: The e2e test suite timed out after exactly 900s (the configured limit) on the MariaDB 8.0 matrix job, with no specific assertion failures — a classic symptom of CI infrastructure slowness or a hanging test rather than a code regression.
Confidence: 0.88

✅ Automatically retrying the workflow

View workflow run

[github-actions]

❌ Test Failure

Analysis: All PG e2e tests consistently time out on STATUS_SETUP across every CI matrix variant, indicating the pg-pg/schema-migration PR likely introduced a regression that causes PostgreSQL flow setup to stall or exceed the 30-second per-test timeout.
Confidence: 0.72

⚠️ This appears to be a real bug - manual intervention needed

View workflow run

[github-actions]

🔄 Flaky Test Detected

Analysis: All 7–8 failures across every matrix shard show "UNEXPECTED STATUS TIMEOUT STATUS_SETUP", meaning CDC workflows timed out during Temporal workflow setup — a classic infrastructure/timing flake, not a logic regression.
Confidence: 0.92

✅ Automatically retrying the workflow

View workflow run

[github-actions]

❌ Test Failure

Analysis: The same set of e2e PostgreSQL tests (TestPeerFlowE2ETestSuitePG) fail at ~31 seconds across all three CI matrix jobs, pointing to a systematic regression — likely a service initialization timeout introduced by the recent docker-compose image tag upgrades — rather than a flaky failure.
Confidence: 0.82

⚠️ This appears to be a real bug - manual intervention needed

View workflow run

[github-actions]

🔄 Flaky Test Detected

Analysis: Three separate e2e matrix jobs failed due to timing/timeout issues: one run hit the 15-minute test suite timeout causing a cascade panic, while the other two had individual WaitFor-based tests fail in different tests across different matrix configurations, with no consistent assertion error pointing to a real code regression.
Confidence: 0.9

✅ Automatically retrying the workflow

View workflow run

[github-actions]

❌ Test Failure

Analysis: The same two PG schema dump tests fail with STATUS_SETUP timeouts across all CI matrix configurations, consistent with a real regression introduced by the recent Docker image tag upgrade (PR #4285) rather than random flakiness.
Confidence: 0.78

⚠️ This appears to be a real bug - manual intervention needed

View workflow run

[github-actions]

🔄 Flaky Test Detected

Analysis: Both failing tests (Test_PG_Schema_Dump_No_Owner_No_Privileges and Test_PG_Schema_Dump_And_CDC) hit "UNEXPECTED STATUS TIMEOUT STATUS_SETUP" across two independent matrix configurations, indicating the Temporal workflow setup phase exceeded the wait deadline due to CI resource contention rather than a code regression.
Confidence: 0.9

✅ Automatically retrying the workflow

View workflow run

[github-actions]

🔄 Flaky Test Detected

Analysis: Both failing tests (Test_PG_Schema_Dump_No_Owner_No_Privileges and Test_PG_Schema_Dump_And_CDC) hit UNEXPECTED STATUS TIMEOUT STATUS_SETUP — a Temporal workflow setup poll timeout — across all CI matrix variants, while the triggering commit only changed unrelated ClickHouse GCS staging code, making this a timing/environment flake rather than a real regression.
Confidence: 0.8

✅ Automatically retrying the workflow

View workflow run

[github-actions]

🔄 Flaky Test Detected

Analysis: All three failing tests hit setup/propagation timeouts ("UNEXPECTED STATUS TIMEOUT STATUS_SETUP", repeated record-count polling mismatches, ClickHouse table-not-found) that are consistent with CI resource contention in a parallelized e2e suite, not a logic regression from the unrelated ClickHouse GCS staging commit.
Confidence: 0.85

✅ Automatically retrying the workflow

View workflow run

[github-actions]

❌ Test Failure

Analysis: CI setup fails deterministically with "E: Unable to locate package postgresql-client-18" because the PostgreSQL apt repository is not configured on the runner, so no tests execute at all.
Confidence: 0.95

⚠️ This appears to be a real bug - manual intervention needed

View workflow run

[github-actions]

🔄 Flaky Test Detected

Analysis: All failures are "UNEXPECTED STATUS TIMEOUT STATUS_SETUP" errors in Temporal-orchestrated e2e tests unrelated to the merged ClickHouse change, consistent with CI resource saturation causing setup phase timeouts.
Confidence: 0.9

✅ Automatically retrying the workflow

View workflow run

[github-actions]

🔄 Flaky Test Detected

Analysis: Both failing tests (Test_PG_Schema_Dump_And_CDC and Test_PG_Schema_Dump_No_Owner_No_Privileges) hit UNEXPECTED STATUS TIMEOUT STATUS_SETUP — a timing failure where the Temporal workflow's setup phase didn't complete within the wait window — which is a classic flaky pattern in e2e tests under CI load, unrelated to the latest ClickHouse staging commit.
Confidence: 0.75

✅ Automatically retrying the workflow

View workflow run

[github-actions]

🔄 Flaky Test Detected

Analysis: The test TestApiMongo/TestCancelTableAddition_NoRemovalAssumed failed with PostgreSQL SQLSTATE 57P01 (administrator_shutdown), meaning a DB connection was terminated by an external admin command mid-query — an infrastructure event unrelated to the test logic.
Confidence: 0.95

✅ Automatically retrying the workflow

View workflow run

[github-actions]

❌ Test Failure

Analysis: Real build failure: e2e/pg_schema_dump_test.go:351 references internal.GetSecondaryPostgresConfigFromEnv which does not exist in the codebase, causing a compilation error.
Confidence: 0.98

⚠️ This appears to be a real bug - manual intervention needed

View workflow run

[claude]

Code Review

Bug found in flow/workflows/setup_flow.go line 316: CreateNormalizedTable skipped even when pg_dump silently does nothing

When the PEERDB_PG_AUTOMATED_SCHEMA_DUMP flag is enabled but the source or destination peer uses SSH tunnels or non-password auth, RunPgDumpSchema silently returns nil without actually running pg_dump.

However, enablePgSchemaDump is still true (set based on the flag alone), so CreateNormalizedTable is also skipped. The result is that no tables are created on the destination -- neither via pg_dump nor via the normal CreateNormalizedTable path.

Suggested fix: Have RunPgDumpSchema return a (bool, error) instead of just error, where the bool indicates whether pg_dump actually ran. Then use that bool (rather than enablePgSchemaDump) as the skipCreateTables argument.

Also checked for CLAUDE.md compliance -- no violations found.

[github-actions]

❌ Test Failure

Analysis: Real bug: TestRunPipeline_ContextCancel fails consistently in all 3 CI matrix jobs with "runPipeline did not return after context cancel", almost certainly caused by the errors.AsType change in commit 55ec870 breaking context-cancellation error detection in the postgres pipeline connector.
Confidence: 0.95

⚠️ This appears to be a real bug - manual intervention needed

View workflow run

[github-actions]

❌ Test Failure

Analysis: Test_PartitionBy fails deterministically across all 4 ClickHouse test suites with the same assertion mismatch (expected "num" but got "(num)" for the partition_key column), indicating a real regression in how partition keys are generated or reported, not a timing/flaky issue.
Confidence: 0.85

⚠️ This appears to be a real bug - manual intervention needed

View workflow run

[github-actions]

❌ Test Failure

Analysis: Test_PartitionBy deterministically fails across all 4 ClickHouse test suites because ClickHouse now returns "(num)" instead of "num" for partition_key in system.tables, indicating a real behavioral regression rather than a flaky failure.
Confidence: 0.88

⚠️ This appears to be a real bug - manual intervention needed

View workflow run

[github-actions]

🔄 Flaky Test Detected

Analysis: All failures trace back to FATAL: terminating connection due to administrator command (SQLSTATE 57P01) — a PostgreSQL catalog connection was killed mid-run by an admin command/shutdown event, which is an infrastructure flake unrelated to any code change.
Confidence: 0.93

✅ Automatically retrying the workflow

View workflow run

[github-actions]

🔄 Flaky Test Detected

Analysis: The e2e test suite hit the hard 900-second timeout (ran 900.644s) with no assertion failures, indicating a slow or hanging test rather than a code regression.
Confidence: 0.92

✅ Automatically retrying the workflow

View workflow run

[github-actions]

🔄 Flaky Test Detected

Analysis: The e2e test package timed out at exactly the 900s hard limit with no assertion failures, indicating a flaky infrastructure/timing issue rather than a code regression.
Confidence: 0.9

✅ Automatically retrying the workflow

View workflow run

[github-actions]

❌ Test Failure

Analysis: All 4 NullEngine test variants fail deterministically at the resync/initial-snapshot stage because data inserted during snapshot doesn't flow through the ClickHouse NullEngine materialized view to the target table, indicating a real bug not a flaky failure.
Confidence: 0.85

⚠️ This appears to be a real bug - manual intervention needed

View workflow run