Add ParadeDB result for c6a.2xlarge (2026-09-18) - #2098
Open
KazukiKandaKK wants to merge 5 commits into
Open
KazukiKandaKK wants to merge 5 commits into
KazukiKandaKK wants to merge 5 commits into
Conversation
Existing ParadeDB results were from 2024-2025 on c6a.4xlarge only. This adds a fresh run on c6a.2xlarge (8 vCPU / 16 GiB, gp3 220GB) using paradedb/paradedb:latest-pg17 via the official benchmark.sh flow, with one manual intervention. The official `benchmark.sh` run failed CREATE INDEX with "insufficient disk space to complete the index build" after COPY 99997497 succeeded. The 220GB root volume was not exhausted by the dataset itself but by leftover intermediate files: the downloaded hits.tsv.gz, the decompressed hits.tsv on the host (~95GB), and the copy of hits.tsv the official `load` script pushes into the container at /tmp/hits.tsv for server-side COPY. `load` only removes these after the index build, so a build that fails mid-way leaves all three on disk at once. Freeing them (host hits.tsv.gz, host hits.tsv, and the in-container /tmp/hits.tsv) restored enough space, and `CREATE INDEX hits_bm25_idx ON hits USING bm25 (WatchID, URL, Title, SearchPhrase, Referer) WITH (key_field='WatchID')` — the same definition as in `load` — was then run manually and verified valid via pg_index.indisvalid. The table itself was untouched throughout; only cleanup of the load-time scratch files was needed. Because the resume path skipped `install`/`load` (already done) and went straight to the 43-query benchmark loop, `load_time` is not available and is recorded as `null` rather than guessed. - Load: COPY 99,997,497 rows (unchanged from the failed attempt) - Data size: 87,979,974,656 bytes - Concurrent QPS: 0.017, error ratio: 0.000
Contributor
|
Results for Logs:
|
Contributor
|
The run of Logs:
|
Contributor
|
Results for Logs:
|
Contributor
|
Results for Logs:
|
Contributor
|
Results for Logs:
|
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Existing ParadeDB results are from 2024-2025 on c6a.4xlarge only. This adds a fresh run on c6a.2xlarge (8 vCPU / 16 GiB, gp3 220GB) using
paradedb/paradedb:latest-pg17via the officialbenchmark.shflow.One manual step was needed. The first run failed
CREATE INDEXwithinsufficient disk space to complete the index buildright afterCOPY 99997497succeeded. The 220GB volume wasn't undersized for the dataset — the failure was leftover load-time scratch files stacking up at once:hits.tsv.gz, the decompressedhits.tsvon the host (~95GB), and the copyloadpushes into the container at/tmp/hits.tsvfor server-sideCOPY.loadonly cleans these up after the index build succeeds, so a build that fails mid-way leaves all three on disk simultaneously. Freeing them (host.gz, host.tsv, in-container/tmp/hits.tsv) recovered enough space, and— the same definition
loaduses — was then run manually and confirmed valid viapg_index.indisvalid. The table and its data were untouched throughout; this was purely disk cleanup, not a different index or a different load.Because the resume path (already-loaded data, index built) skipped straight to the 43-query loop instead of going through
install/loadagain,load_timeisn't available from this run and is recorded asnullrather than a guess.COPY 99,997,497rowsHappy to add a note to
paradedb/loadabout cleaning up/tmp/hits.tsv(or the host tsv) before the index build if that failure mode is worth guarding against for future runs — let me know if you'd rather see that as a separate PR.