Skip to content

Add methods to check if the query requires traces/statediffs NET-433 - #67

Merged
kalabukdima merged 1 commit into
masterfrom
traces-detection
May 7, 2026
Merged

Add methods to check if the query requires traces/statediffs NET-433#67
kalabukdima merged 1 commit into
masterfrom
traces-detection

Conversation

@kalabukdima

Copy link
Copy Markdown
Contributor

It can be used to route queries to traceless versions of the datasets in the portal

@kalabukdima
kalabukdima requested a review from tmcgroul May 6, 2026 10:57
@kalabukdima kalabukdima self-assigned this May 6, 2026
@tmcgroul

tmcgroul commented May 6, 2026

Copy link
Copy Markdown
Contributor

what problem does it solve? i think it's not smth common for us to have both tracefull and traceless at the same time

@kalabukdima

kalabukdima commented May 6, 2026

Copy link
Copy Markdown
Contributor Author

Please see the description at https://linear.app/sqd-ai/issue/NET-433/support-traceless-hotblocks

I see it as the simplest option to provide better latency for EVM datasets. If it helps, we can think about a storage-level redesign to avoid duplication

See also: subsquid/sqd-portal#100

@kalabukdima
kalabukdima merged commit 849f6ea into master May 7, 2026
1 check passed
@kalabukdima
kalabukdima deleted the traces-detection branch May 7, 2026 10:34
dzhelezov added a commit to dzhelezov/data that referenced this pull request Aug 11, 2026
…trands, honest retry counting, replay-safe finality log

F67-3: perform_dataset_compaction now abandons the tables it already
materialized when the transaction does not commit its chunk (Canceled or
retry-exhausted), mirroring WriteController::abandon_tables on the ingest
path. A lost compaction race previously stranded an orphan DIRTY_TABLES
marker that pins the reclaim watermark until the next restart, one per lost
race while hotblocks re-attempts compaction every minute. Adds a racing-
compaction orphan-probe test (fails pre-fix: one orphan dirty marker).

F67-4: count a retry restart only once a replacement transaction is actually
created. A run terminated by the post-backoff deadline check never re-ran the
callback, so counting it earlier inflated restart/attempt telemetry with
attempts that never happened.

F67-5(b): racing_writers bars both writers behind a barrier and asserts a
real restart delta, so "consistent under contention" cannot pass vacuously on
a run that never actually contended.

F67-7: the "finalized head was ignored" decision is returned from the
retryable new_chunk closure and logged once post-commit, instead of being
emitted inside the closure where an optimistic-transaction replay duplicates
it once per conflict attempt.

Co-Authored-By: Claude Opus 4.8 <[email protected]>
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.

2 participants