deepakpanda93 commented on issue #9244:
URL: https://github.com/apache/hudi/issues/9244#issuecomment-4845368525

   Closing this out — it's a Flink write-failure-recovery scenario, and the 
reader-side gap it exposed is addressed in 1.x.
   
   What happened: the datanode downtime killed the Flink write mid-flight and 
left an uncommitted / un-rolled-back log block in the bucket file group; after 
restart, that leftover block was read alongside the retried write, producing 
duplicates that carry the failed write's hoodie meta fields.
   
   In 1.x this is handled:
   
   - the log reader now skips uncommitted log files/blocks (HUDI-6952, 
[#9879](https://github.com/apache/hudi/pull/9879), in 1.0.0+), so a failed 
write's leftover block is no longer read;
   - the completion-time-based timeline (1.0) means readers only surface data 
from completed instants;
   - Flink failover handling was hardened (partial failover [#7208], plus fixes 
to re-trigger failover after pending instants are recommitted).
   
   The original report was on 0.13.0, which predates these. Since the timeline 
for the failed run was never shared and the relevant recovery/reader code has 
been substantially reworked, I'm closing this. If you can still reproduce 
duplicates after a write failure on a recent 1.x release (Flink + bucket index 
+ MOR), please reopen with the .hoodie/ timeline listing around the failure and 
we'll dig in. Thanks!


-- 
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]

Reply via email to