JackieTien97 opened a new pull request, #18920:
URL: https://github.com/apache/iotdb/pull/18920

   ## Description
   
   Fixes #18633.
   
   `SHOW AVAILABLE URLS` currently includes `Unknown` and `Stopped` DataNodes. 
Java SDK auto-discovery puts these addresses into the normal query rotation, so 
a persistently unreachable node can repeatedly add a connection timeout to 
otherwise successful queries. This change filters offline nodes on the server 
and preserves the client's initial addresses as recovery seeds.
   
   ### Rolling restart scenario
   
   A client can discover only nodes 2 and 3 while node 1 is stopped. Node 1 may 
recover before the next successful discovery refresh, while subsequent 
connection attempts to the cached nodes fail during maintenance. Filtering the 
server response alone leaves both business reconnection and background 
discovery dependent on the same stale list, even though node 1 is reachable 
again.
   
   The client now tries the discovered endpoints first, then falls back to the 
initial `nodeUrls`. If node 1 was configured as a seed, it remains a recovery 
candidate after being omitted from a discovery snapshot. A successful business 
reconnection through a seed requests a prompt background refresh so recovered 
nodes can re-enter normal routing.
   
   ### Implementation
   
   - Exclude `Unknown` and `Stopped` from `SHOW AVAILABLE URLS`, continue 
excluding `Removing`, and retain `Running` and `ReadOnly` nodes.
   - Preserve independent copies of the initial addresses in 
Session/SessionPool and the node supplier, including single-host configuration. 
Discovery updates and changes to the default endpoint do not replace the 
recovery seeds.
   - Apply discovered-first, seed-second recovery to both 
`SessionConnection.reconnect()` and `NodesSupplier.run()`. Deduplicate 
addresses within each attempt round and retain the existing bounded business 
reconnection rounds.
   - Let background discovery try other discovered nodes and seeds in the same 
run after its current connection fails. Coalesce refresh requests from sessions 
sharing a supplier.
   - Publish successful empty discovery results while retaining seeds. Select 
query endpoints from one snapshot so a concurrent refresh to an empty list 
cannot invalidate selection. Seeds are not merged into the normal query 
rotation.
   
   Recovery still requires a reachable configured seed when all discovered 
endpoints fail. The seed fallback requires the updated Java SDK. This change 
does not add endpoint backoff for nodes that the server reports as `Running` 
but that are unreachable from a particular client.
   
   ### Validation
   
   Validated on JDK 17:
   
   - **239 targeted unit tests passed:** 237 session/client tests and 2 
`ShowAvailableUrlsTaskTest` tests. Coverage includes seed recovery, 
discovered-node priority, empty snapshots, duplicate addresses, bounded 
retries, caller mutation, refresh coalescing, and concurrent query selection 
during refresh.
   - **2 `SessionSeedNodeFallbackIT` tests passed on a real 1 ConfigNode / 3 
DataNode cluster:**
     - `testQueriesDuringRollingRestart`: Session and SessionPool queries 
return the expected data before and after each DataNode is stopped and 
restarted; discovery excludes stopped nodes and restores recovered nodes.
     - `testBusinessAndBackgroundRecoveryThroughOmittedSeed`: freeze periodic 
discovery, recover an omitted seed, make the two cached endpoints unavailable 
(`Unknown` and `Stopped`), and verify business reconnection with a metadata 
query and independent background discovery recovery through the seed.
   - Full-reactor `mvn test-compile -DskipTests` and `mvn test-compile 
-DskipTests -P with-zh-locale` passed.
   - Spotless and Git whitespace checks passed.
   
   Validation notes: RAT remained enabled; a local excludes file omitted five 
pre-existing untracked SVG files outside this PR. The rolling-restart test 
tolerates `IoTDBConnectionException` only when closing stale sessions during 
cleanup. Query and discovery failures are not ignored; production close 
behavior is unchanged.
   
   <hr>
   
   This PR has:
   
   - [x] been self-reviewed.
   - [x] added comments explaining recovery ordering and snapshot handling.
   - [x] added unit tests covering the new paths.
   - [x] added integration tests.
   - [x] been tested in a test IoTDB cluster.
   
   ##### Key changed/added classes
   
   `ShowAvailableUrlsTask`, `INodeSupplier`, `NodesSupplier`, 
`SessionConnection`, `Session`, `SessionPool`, `ShowAvailableUrlsTaskTest`, 
`SessionSeedNodeFallbackTest`, and `SessionSeedNodeFallbackIT`.
   


-- 
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