Follow the applied assignment type for schema sources - #77
Merged
Merged
Conversation
Two changes to how the worker sources query-engine schemas. The CDN manifest loop now follows the assignment in force instead of polling unconditionally. The applier publishes whether the applied assignment resolves chunk schemas by dataset type: true from startup and under legacy, false once a split assignment applies, true again when a legacy one does. While false the manifest is not fetched at all and any retry backoff is cut short; on resume the manifest is refreshed at once rather than at the next tick. What the type registry loaded before a pause stays in memory, so chunks held from an earlier assignment still resolve by type under split if their type had loaded. A schema id republished with different contents now replaces the stored copy instead of refusing the whole bundle as a permanent fault. The bundle in force is the meaning of its ids; an identical file is still left where it is, and the merge log reports how many ids were replaced. The p2p controller builds the applier once so both loops share it, and the e2e harness is wired the same way. README, IB-44, IB-44b, FM-53, FM-53b, P-SCHEMA-REFRESH and ADR-23 are updated; ADR-23 carries a dated revision note. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
define-null
force-pushed
the
schema-source-follows-assignment
branch
from
September 8, 2026 14:05
2d8bde2 to
4cf9ec1
Compare
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Why
After the engine bump in #76, the CDN's old schema layout fails to parse, while the scheduler's bundle already uses the new layout. Workers continued polling the CDN even after applying a split assignment whose chunks resolve schemas by ID.
The bundle loader also rejected any existing schema ID republished with different bytes. This made workers with cached schemas reject updates that fresh workers accepted.
What changes
The assignment applier owns a concurrent CDN refresh task:
CDN fetch or parse failures do not fail or block legacy assignment application. The task retains previously loaded schemas and retries with exponential backoff. Before the first successful load, dynamic queries requiring type schemas return
server_error. Split assignments still require their separate schema bundle to download, validate, and install successfully.The watch channel and separate CDN subsystem are removed. The applier observes refresh-task failures, and dropping it aborts any remaining refresh task. The end-to-end harness uses the same ownership model.
A bundle that republishes an existing schema ID with different contents replaces its stored file and loaded schema. Identical files remain untouched, and IDs absent from the new bundle remain in the accumulated store.
Validation
cargo test --locked --offline --all-targets.cargo fmt --checkandgit diff --checkpassed. Clippy reports only the existing warning inpolars_target.rs.README and the relevant schema-source specifications are updated.
Remaining behavior