alpass163 opened a new pull request, #18800:
URL: https://github.com/apache/iotdb/pull/18800

   ## Problem
   
   A rename is pre-released to every DataNode (PRE_UPDATE_TABLE) before the 
ConfigNode
   commits it. A DataNode that registers during that phase is not in the 
broadcast target
   set — it is frozen when the propagator is built — but it does receive the 
later commit
   release. Such a DataNode ends up in a wrong state:
   
   - its activation snapshot still holds the table under the old name, and
   - `commitUpdateTable` finds no pre-released entry for the new name, so it 
only logs
     "skip commit update ... because it has been handled" and returns.
   
   The old name therefore stays in the cache as a usable table, and the new 
name is
   reported as TABLE_NOT_EXISTS on that DataNode. It only heals on lease 
recovery or
   restart.
   
   ## Fix
   
   - **DataNode** (`DataNodeTableCache`): on a commit that finds no 
pre-released entry and
     carries an `oldName`, repair the cache — drop the old name from 
`databaseTableMap` and
     leave a `NonCommittableTsTable` placeholder for the new name. The 
placeholder routes
     the next lookup of the new name into the existing lazy-fetch path, which 
pulls the
     committed definition from the ConfigNode. A repeated commit is idempotent: 
while the
     placeholder exists the commit returns at the `NonCommittableTsTable` 
check, and after
     a fetch has consumed it the repair skips re-marking a table that is 
already cached.
   - **ConfigNode** (`AbstractAlterOrDropTableProcedure`): the commit release 
of a rename
     is now broadcast through `ClusterCachePropagator`, the same delivery 
guarantee the
     pre-release already has. 


-- 
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