Skip to content
This repository was archived by the owner on Nov 21, 2026. It is now read-only.

benchmarks: native time-series arms use the engine's default SHARDS (#241) - #242

Merged
tae898 merged 1 commit into
mainfrom
ts-shards-engine-default
Oct 6, 2026
Merged

tae898 merged 1 commit into
mainfrom
ts-shards-engine-default

Conversation

@tae898

@tae898 tae898 commented Oct 6, 2026

Copy link
Copy Markdown

Closes #241.

The native time-series arms (embedded and served) no longer fix SHARDS 4: the clause is left out, so the engine derives the shard count from its async workers, which is what upstream names (ArcadeData#9166 and the time-series docs). A round 2 best-practice probe measured the fixed 4 at 0.505 ms against 1.007 ms for the default 11 on the 12 h hourly aggregate (ratio of medians 1.99, five runs per arm, JDK 25, official 26.10.1 jars, same answers), so the default is the setting that does not flatter ArcadeDB.

TS_SHARDS=4 reproduces the old arm. Rows record ts_shards (0 for the default) and ts_shard_count_effective (read from schema:types). PROTOCOL.md and CAMPAIGN.md row 54 are updated, and the tests assert the DDL has no SHARDS clause by default, that the env var brings it back, and that the readback parses the engine report. I also created a type without the clause through the 26.10.1 wheel to confirm the DDL parses and the report carries shardCount (3 under cores 4-7).

Not verified: JDK 21, and the effect on mini (the mini rows are the measurement). The time-series stage has not started on mini, so no rows are split.

🤖 Generated with Claude Code

https://claude.ai/code/session_01Ubz3m6QXHESerCkehC1qgi

…DS, not a fixed 4 that flatters ArcadeDB (#241)

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Ubz3m6QXHESerCkehC1qgi
@tae898
tae898 merged commit e2e669d into main Oct 6, 2026
2 checks passed
@tae898
tae898 deleted the ts-shards-engine-default branch October 6, 2026 17:53
Sign up for free to subscribe to this conversation on GitHub. Already have an account? Sign in.

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Native time-series arms: use the engine's default SHARDS, not a fixed 4 that flatters ArcadeDB

1 participant