PDGGK commented on PR #18725:
URL: https://github.com/apache/iotdb/pull/18725#issuecomment-6581407239

   In case it helps with the IT runs: while testing this PR at 87b58e81, 
`IoTDBLTTBTableFunctionIT` failed intermittently with `719: There is not enough 
memory to execute current fragment instance`, in 9 of 12 runs with 
`-PTableSimpleIT`.
   
   The main cause is outside this PR. On master, when the consumer closes a 
local channel first, the producer's fragment instance stays in `FLUSHING` and 
keeps its memory for about a second, until `SimpleQueryTerminator` finishes it. 
An LTTB root FI under an `ORDER BY` reserves about 6 MB of the 16.09 MB 
operator pool of the IT DataNode, so one leftover FI from the previous query is 
enough to reject the next one. I opened #19084 for that race. With it applied 
on 87b58e81, the class passed 8 runs in a row with `-PTableSimpleIT`.
   
   `-PTableClusterIT` still failed with #19084 applied: 6 and 5 of the 15 tests 
erred with 719 in two runs, against 9 and 8 without it. That profile sends each 
query to all three DataNodes at once, so three copies of a single-region LTTB 
query need about 3 × 6 MB on the DataNode that holds the region, more than its 
16.09 MB pool. This part does depend on the size of the LTTB reservation. A 
smaller `lttb_buffer_size_in_bytes` for the IT would probably avoid it, but 
`MppCommonConfig` has no setter for that key yet, and I haven't tried it.
   
   I haven't rerun either profile on 640d4020.
   


-- 
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