d-wang-commit opened a new pull request, #19151:
URL: https://github.com/apache/iotdb/pull/19151

   ## Description
   
   Follow-up to #19126. A DataNode in `ReadOnly(DiskFull)` can receive a new 
writable directory through a configuration reload. Reset disk protection once 
after a successful explicit reload so subsequent operations use the current 
directory configuration.
   
   Both `LOAD CONFIGURATION` and a nonempty `SET CONFIGURATION` record whether 
the DataNode was `ReadOnly(DiskFull)` before applying the configuration. After 
success, they set the node to `Running` and clear the old recovery checks only 
if the status and reason are still exactly `ReadOnly(DiskFull)`. This preserves 
a disk-full condition first raised during the reload, a concurrent change to a 
higher-priority reason, and protection when the reload fails.
   
   Remove the tier-specific callbacks introduced in #19126 for looking up 
failed groups across directory reinitialization. Construction and selection 
failures register their recovery conditions directly, using the candidate paths 
or the folder manager as the key. Heartbeats still require healthy aggregate 
space and a passing check for every recorded source, and probe each candidate 
without retry sleeps.
   
   ### Tests
   
   - Added `DiskFullHotReloadTest` for successful and failed reloads, empty 
requests, clearing old checks and preserving other read-only reasons.
   - On the previously tested implementation, 78 focused unit tests passed and 
4 existing tests were skipped. All 12 executions of the existing configuration 
and system-status SQL integration tests passed across the one- and 
three-DataNode topologies. Four isolated runs exercised status changes during 
both reload entry points.
   - In the three-replica simulation, the target directory reported zero free 
bytes while the aggregate ratio remained around 93%. The DataNode stayed 
read-only across heartbeats until space was released. Repeated TsFile creation 
errors during consensus replay led to `UnrecoverableError`, which persisted 
after space was released.
   - On this follow-up commit, the clean Chinese-locale reactor test 
compilation passed. Git tree comparison verified that the follow-up on top of 
#19126 exactly reproduces that tested implementation plus the intervening 
upstream changes. The full suite was not rerun to completion on this commit.
   
   ### Side effects and risks
   
   Any successful explicit reload can reset an existing disk-full state, 
including a reload of an unrelated setting. If a disk remains full, the node 
can show `Running` until another operation detects it. Access to tier folder 
managers after all candidates fail initialization is a separate issue.
   
   <hr>
   
   This PR has:
   - [x] been self-reviewed.
   - [x] added comments explaining non-obvious behavior.
   - [x] added unit tests for the new paths.
   
   <hr>
   
   ##### Key changed/added classes (or packages if there are too many classes) 
in this PR
   
   - `IoTDBDescriptor`, `StorageEngine`, and `CommonConfig` implement the 
guarded reset.
   - `TierManager`, `TsFileNameGenerator`, `DirectoryStrategy`, and 
`FolderManager` remove the tier-specific routing of recovery checks.
   - `DiskFullHotReloadTest` exercises the reload entry points.
   


-- 
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