mongodb: fix leaked changestream cursor

[jgao54]

Fixes a changestream cursor resource leak that can get triggered when the context is already cancelled. ChangeStream.Close executes a killCursors command against the server; but skips it if context is invalid. This has led to the error "Cannot open a new cursor since too many cursors are already opened" for DocumentDB instances that cap cursor limit. Fix is to always use a fresh context.

Also fixes leak caused by ignoring reassignment of changestream when recreateChangeStream is called. Without the closure, defer captures the original changestream at time it was evaluated.

Fixes: DBI-902

[claude]

Code review

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

[codecov]

❌ 2 Tests Failed:

Tests completed	Failed	Passed	Skipped
3021	2	3019	391

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

github.com/PeerDB-io/peerdb/flow/e2e::TestPeerFlowE2ETestSuiteMySQL_CH

Stack Traces | 0.02s run time

=== RUN   TestPeerFlowE2ETestSuiteMySQL_CH
=== PAUSE TestPeerFlowE2ETestSuiteMySQL_CH
=== CONT  TestPeerFlowE2ETestSuiteMySQL_CH
2026/07/10 19:26:21 INFO Received AWS credentials from peer for connector: ci x-peerdb-additional-metadata={Operation:FLOW_OPERATION_UNKNOWN}
--- FAIL: TestPeerFlowE2ETestSuiteMySQL_CH (0.02s)
2026/07/10 19:39:59 INFO Received AWS credentials from peer for connector: ci x-peerdb-additional-metadata={Operation:FLOW_OPERATION_UNKNOWN}
2026/07/10 19:39:59 INFO Received AWS credentials from peer for connector: clickhouse x-peerdb-additional-metadata={Operation:FLOW_OPERATION_UNKNOWN}

github.com/PeerDB-io/peerdb/flow/e2e::TestPeerFlowE2ETestSuiteMySQL_CH/Test_MySQL_BinlogIncident

Stack Traces | 58s run time

=== RUN   TestPeerFlowE2ETestSuiteMySQL_CH/Test_MySQL_BinlogIncident
=== PAUSE TestPeerFlowE2ETestSuiteMySQL_CH/Test_MySQL_BinlogIncident
=== CONT  TestPeerFlowE2ETestSuiteMySQL_CH/Test_MySQL_BinlogIncident
2026/07/10 19:39:01 INFO Received AWS credentials from peer for connector: ci x-peerdb-additional-metadata={Operation:FLOW_OPERATION_UNKNOWN}
2026/07/10 19:39:01 INFO Received AWS credentials from peer for connector: clickhouse x-peerdb-additional-metadata={Operation:FLOW_OPERATION_UNKNOWN}
2026/07/10 19:39:01 INFO fetched schema x-peerdb-additional-metadata={Operation:FLOW_OPERATION_UNKNOWN} table=e2e_test_mych_8nq80ex6.test_nullengine
    clickhouse_mysql_test.go:2157: 
        	Error Trace:	.../flow/e2e/mysql.go:130
        	            				.../flow/e2e/clickhouse_mysql_test.go:2157
        	            				.../hostedtoolcache/go/1.26.4.../src/runtime/asm_amd64.s:1771
        	Error:      	Received unexpected error:
        	            	MySQL execute error: readInitialHandshake: io.ReadFull(header) failed. err EOF: connection was bad
        	Test:       	TestPeerFlowE2ETestSuiteMySQL_CH/Test_MySQL_BinlogIncident
2026/07/10 19:39:29 INFO fetched schema x-peerdb-additional-metadata={Operation:FLOW_OPERATION_UNKNOWN} table=e2e_test_mych_5o3mc7sx.test_unsigned
--- FAIL: TestPeerFlowE2ETestSuiteMySQL_CH/Test_MySQL_BinlogIncident (57.98s)

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: Test_MySQL_BinlogIncident failed on a transient MySQL handshake EOF ("readInitialHandshake: io.ReadFull(header) failed. err EOF: connection was bad") while connecting to its own throwaway debug MySQL container during setup — a connection/startup-timing race, not a logic bug.
Confidence: 0.9

✅ Automatically retrying the workflow

View workflow run