chihsuan commented on PR #10540: URL: https://github.com/apache/ozone/pull/10540#issuecomment-5306535440
Hi @jojochuang š I looked into this more deeply and worked through the design alternatives with Claude Code. The original implementation had no feature flag, so I added the datanode config to reduce rollout risk. However, it only gates the DN side. SCM still maintains the command index unconditionally. To make this easier to review and roll out, Iām considering splitting it into two steps: 1. DN terminal-outcome reporting and SCM metrics only, without pending-op mutation, indexing, or recipient tracking. 2. Proactive pending-ADD clearing, behind a reconfigurable SCM flag, after validating the failure rate and retry/backoff semantics. I also noticed that the Jira workaround of setting `hdds.scm.replication.event.timeout=1m` is invalid with the default 6-minute datanode offset. Reducing both may also expire queued or slow replication too aggressively. I found a safer interim mitigation: increase the global in-flight factor to 1.0, or temporarily disable the global limit with monitoring. If this direction makes sense, Iām happy to close this PR and open smaller, focused PRs. I can also submit the design as a Markdown document for review. Iād appreciate your guidance before proceeding. Thanks! š -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
