JackieTien97 opened a new pull request, #18587:
URL: https://github.com/apache/iotdb/pull/18587

   ## Description
   
   Edge still creates query scheduling notifications through an unbounded 
cached pool, while fragment dispatch uses a hard-coded minimum of 20 workers. 
Add restart-only settings and set both Edge defaults to 2:
   
   | Setting | General default | Edge default |
   | --- | --- | --- |
   | `driver_task_scheduler_notification_thread_count` | 0: retain the existing 
unbounded cached pool | 2 |
   | `fragment_instance_dispatch_thread_count` | 0: use max(20, available 
processors × 2) | 2 |
   
   Positive notification sizes queue callbacks on a limited worker pool; 
workers can exit after 60 seconds of inactivity. Callbacks do not fall back to 
the threads completing driver futures. Fragment dispatch retains its existing 
caller-runs behavior when the pool is busy. Negative configuration values are 
rejected.
   
   Remove the unused BinaryAllocator implementation, its configuration, metrics 
and lifecycle hooks. Repository-wide call-site inspection found 
allocation/deallocation calls only in its unit tests; production references 
merely started it through GC monitoring or configuration reload. The old 
auto-releaser could remain blocked in `ReferenceQueue.remove()` after shutdown, 
accumulating threads across restarts. Removing the unused subsystem prevents 
those workers from being created. Normal BLOB handling remains covered by the 
real Edge read/write test.
   
   ConfigNode and SchemaRegion continue to use Ratis; DataRegion continues to 
use IoTConsensus.
   
   ## Validation
   
   - Clean distribution build; all six distribution artifacts produced. The 
packaged node-commons jar contains no binaryallocator classes.
   - Four unit tests passed: configuration 
defaults/validation/startup/restart-only behavior, driver scheduling, and 12 
blocked-driver notifications completing through two workers without executing 
on the completing thread.
   - Seven `IoTDBEdgeBasicIT` tests passed, including tree/table read/write, 
concurrent table queries, 40 cross-database join queries across four concurrent 
connections, consensus assertions, and a 64 KiB BLOB verified before/after 
flush and configuration reload.
   - Edge starts and reloads a configuration retaining the legacy 
`enable_binary_allocator=true` key. Thread dumps show no BinaryAllocator 
worker. The cross-database join dump shows two fragment-dispatch workers and 
two driver-notification workers.
   - Full-reactor English and Chinese test compilation passed. Locally built 
node ZIP dependencies were installed first because this branch version is not 
available in the remote snapshot repository.
   - Linux aarch64 with a two-CPU quota and 4 GiB memory: 100 devices × 100 
DOUBLE fields at 1 Hz, two writers and four readers for 180 seconds. All 
1,810,000 written values were verified before and after explicit FLUSH; no 
request failures or missed write rounds. The notification pool had two workers 
during the workload and zero after a subsequent 75-second idle interval. No 
BinaryAllocator workers appeared.
   


-- 
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