Skip to content

Delta apply: log why nothing was taken over; cluster results (#49) - #50

Merged
gspark-etri merged 1 commit into
mainfrom
feat/rebase-log
Oct 22, 2026
Merged

gspark-etri merged 1 commit into
mainfrom
feat/rebase-log

Conversation

@gspark-etri

Copy link
Copy Markdown
Contributor

Background

In the #41 cluster check, a prefill pod did not take over the idle decode artifact. That was expected, but the log could not show why (#49).

Changes

  • PullStats.BaseBest: the bytes the best offered base keeps at the same offsets, whether it was taken or not. chooseBase now returns its best candidate together with the decision.
  • The materializer logs when nothing is taken over:
    • no artifact to take over (k in use, j used within <rebase-idle>);
    • took over nothing: of n idle artifacts the best keeps x of y B at the same offsets (p %, 50 % needed).
  • busy is split out of idle.
  • deploy/csi/README.md: the Delta apply: turn an idle artifact on the node into the next variant, writing only what changed #41 cluster results:
    • a decode B8 pod was Running 12 s after creation, taking over the idle B2 artifact (15.0 GB left in place, 28.3 MB read, 7.5 s pull, byte-identical, read-only), against about a minute before;
    • an artifact in use was not touched;
    • a prefill pod took over nothing.

Verification

step pod Running after pull check
decode B2, cold 160 s 10.20 GB from the store identical
decode B8, B2 idle for 1 min 12 s took over B2: 15.0 GB in place, 28.3 MB read, 7.5 s sha256 identical; read-only
decode B2 while the B8 pod runs 69 s no takeover (B8 in use); 10.20 GB from the node cache identical
prefill C128, B8 idle 82 s no takeover; 1.9 GB store + 7.5 GB cache identical

Cleanup was verified: no mounts, leases or files left, and the CSIDriver is deleted.

Compatibility and limits

  • A new stats field and new log lines only.

Related Issue

Closes #49

🤖 Generated with Claude Code

#49)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@gspark-etri
gspark-etri merged commit d2a8c56 into main Oct 22, 2026
3 checks passed
@gspark-etri
gspark-etri deleted the feat/rebase-log branch October 22, 2026 05:12
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Delta apply: log why no artifact was taken over

1 participant