andygrove commented on PR #2293:
URL:
https://github.com/apache/datafusion-ballista/pull/2293#issuecomment-5286920962
Good check — I just measured this on #2241.
For the Rust CI matrix, git-ref pinning is roughly neutral. Comparing latest
main (DF 54 from crates.io) against `datafusion-55` (git tag 55.0.0-rc2):
| Job | main | datafusion-55 | Δ |
|---|---:|---:|---:|
| Ballista Rust | 32m 51s | 26m 35s | −6m |
| h2o | 29m 39s | 27m 40s | −2m |
| TPC-H SF10 | 35m 07s | 27m 07s | −8m |
| TPC-DS SF1 | 30m 23s | 29m 58s | −25s |
| Docker | **20m 27s** | **26m 27s** | **+6m** |
| Ballista Web TUI | 3m 00s | 1m 08s | −2m |
| Dev | 53s | 37s | −16s |
Two caveats: this is DF 54 → DF 55, not just source-type — most of the wins
on the Rust jobs are probably upstream code-size differences, not
git-vs-registry. And there's real runner noise; Docker on sibling DF 55
branches came in at 26m, 30m, and 33m on the same rev.
So my read is:
- **Rust matrix jobs**: no measurable git-ref penalty. `rust-cache` warms
after the first fetch, and the compile graph drives the rest.
- **Docker job**: ~+5–10 min consistently. Worth a mention in the doc so
contributors aren't surprised. Once DF 55 GA lands on crates.io and we can
switch back to a version pin, that should recover.
(FWIW I also tested `[patch.crates-io]` vs `{ git = ... }` in
`workspace.dependencies` — `Cargo.lock` comes out byte-identical either way, so
the pin *syntax* doesn't matter; only the underlying source type does.)
--
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]
