yuqi1129 opened a new issue, #13181:
URL: https://github.com/apache/gravitino/issues/13181

   ### Describe the subtask
   
   Row OCC makes writes safe; it says nothing about what a reader sees. Reads 
may be served from the per-node cache that `EntityChangeLogPoller` invalidates 
every `gravitino.entityChangeLog.pollIntervalSecs` (default 3 s) — eventual, 
not a bound. Some aggregate reads span statements (`getTableByIdentifier` reads 
the row, then the columns), so a drop or overwrite in between returns a row 
with no columns.
   
   Proposed change:
   - Write the read contract into `docs/`: which paths may return a stale 
cached entity, which must read through (in-use checks, owner lookup for 
permission checks, lifecycle checks before a drop).
   - Enforce read-through on those paths (or route them to the shared cache 
when one is configured).
   - Root-version re-validation for row + columns / row + versions reads: 
re-read the root after the child read, retry once on mismatch.
   - Tests: read routed to the other server before invalidation; drop between 
root and child reads.
   
   Design: implementation design Part 5. Gate G5.
   
   ### Parent issue
   
   https://github.com/apache/gravitino/issues/10238


-- 
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