andygrove opened a new pull request, #2171:
URL: https://github.com/apache/datafusion-ballista/pull/2171

   ## Summary
   
   Follow-up to #2161. Refresh the SF1000 numbers on 
`docs/source/contributors-guide/benchmarking.md` with results from a run that 
mounts a **dedicated 1000 GiB gp3 EBS PVC per executor** at `/data` and passes 
`--work-dir /data` to the executor. This isolates Ballista's shuffle writes 
from the container overlay FS / shared root EBS volume — matching the 
`spark.kubernetes.executor.volumes.persistentVolumeClaim.spark-local-dir-1` 
pattern that Spark already uses on the same cluster.
   
   ## Impact
   
   Same commit (`main` @ `696ca29b`), same cluster shape, only difference is 
the dedicated shuffle PVC:
   
   | Query | Prior run | With PVC | Speedup |
   |---:|---:|---:|---:|
   | Q8 | 189.30 s | **45.74 s** | 4.1× |
   | Q9 | 154.88 s | 67.27 s | 2.3× |
   | Q17 | 60.46 s | 47.29 s | 1.3× |
   | **Q18** | **282.79 s** | **78.32 s** | **3.6×** |
   | Q20 | 88.93 s | 97.13 s | ~parity |
   | **19Q comparable subset** | **1080 s** | **722 s** | **1.5×** |
   
   Q8 in-suite (45.74 s) is now within 1 s of Q8 standalone (47.4 s in a solo 
run), confirming the pre-PVC slowdown was root-EBS bandwidth contention from 
prior queries' dirty pages, not anything Ballista-internal — see the analysis 
in [#2168](https://github.com/apache/datafusion-ballista/issues/2168).
   
   ## What changed in the doc
   
   - **Environment:** record the underlying `r6i.24xlarge` node type (96 vCPU, 
768 GiB, EBS-only) and the new dedicated Ballista PVC.
   - **New "Executor storage" section:** why the dedicated volume matters, with 
a short `EXPLAIN ANALYZE`-based explanation of the pre-PVC contention.
   - **Ballista configuration:** add `--work-dir` and 
`--grpc-server-max-*-message-size` rows; note that the 16 MiB gRPC default 
currently causes Q11/Q21/Q22 to fail.
   - **Results:** refreshed Ballista column; mark Q11/Q21/Q22 as `FAIL` pending 
investigation of the client-submission channel's 16 MiB ceiling; publish Q1–Q10 
+ Q12–Q20 subset totals so cross-engine comparison is on the same set of 
queries.
   - **Reproducing:** updated `ballista-executor` invocation with the new flags.
   
   ## Test plan
   
   - [x] Doc renders as GitHub markdown (tables + code blocks).
   - [x] Prettier clean.
   - [ ] Reviewer sanity-check that the results numbers match the referenced 
result JSON (available on request).
   


-- 
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]


---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to