<@U04S8PLJWMT> i am using v3.50.5 of mysql source ...
# replication-ask-ai
m
@kapa.ai i am using v3.50.5 of mysql source connector and v3.6.35 of postgres source connector. are these connectors versions are stable? is there any issue in these versions?
k
Connector Version Stability Assessment MySQL Source — v3.50.5 This version has a known critical bug. There is a confirmed
ArithmeticException: / by zero
error in
MySqlSourceJdbcPartitionFactory.internalCalculateBoundaries
that affects versions 3.50.2, 3.50.3, 3.50.5, and 3.50.6. [MySQL ArithmeticException issue] [Regression tracking issue] Impact: • Blocks initial syncs on larger tables (reported on tables ~100 GB and ~942 MB) • Occurs during the partitioning stage with incremental | append + deduped sync mode and CDC (binlog) replication • Smaller tables in the same database may sync without issues Fix: A fix was shipped in v3.0.8 (as noted in the issue comments). A later patch bump to v3.52.4 also addressed a related CDC payload-limit issue. [Fix confirmation] [CDC payload fix] Recommendation: Upgrade away from v3.50.5. If you cannot upgrade, the reported workaround is to downgrade to v3.11.21 (Docker Hub tag), though this is not ideal for Cloud users. Postgres Source — v3.6.35 This version also has known issues: 1. CDC LSN advancement bug — Versions from ~3.6.19 onward (including 3.6.35) use a PgJDBC version (
42.7.x
) that automatically advances
confirmed_flush_lsn
, causing CDC syncs to fail with: "Saved offset is before replication slot's confirmed lsn" Versions 3.6.15 and 3.6.18 are reported as stable workarounds. [LSN issue] [PgJDBC issue] 2.
IndexOutOfBoundsException
during CDC sync
— Reported on v3.6.35 after several hours of running, causing unreliable CDC replication. [IndexOutOfBounds issue] 3.
op:m
message events not filtered in shutdown cleanup loop
— Affects v3.6.35+ and can cause
IllegalStateException: ChangeEvent contains no before or after records
during Debezium shutdown. [op:m filter issue] 4. Write timeout if destination is slow — Also reported on v3.6.35 when the destination (e.g., Databricks) takes time to initialize. [Timeout issue] Recommendation: v3.6.35 is not recommended for production CDC workloads. Consider using v3.6.15 or v3.6.18 as a stable fallback for CDC, or check the latest available version for fixes. Summary It is strongly recommended to upgrade both connectors to their latest available versions or use the known-stable fallback versions mentioned above. 👍 0 button 👎 0 button ℹ️ Tip: Mention @kapa.ai in the thread for follow-ups or DM the app for 1-to-1 help., with interactive elements