fix: skip needless remote connects in local read paths - #115
Closed
lehuan5062 wants to merge 10 commits into
Closed
Conversation
Add 5-second connect timeouts to gRPC and QUIC transports, and time-box the remote branch resolution in status() to 3 seconds. When the configured remote is unreachable, status calls now degrade gracefully instead of stalling for the full idle timeout (~30s for QUIC, 10s+ for retries). Critical fixes: 1. Time-box the entire remote resolution operation (both repository.remote() and branch::load_remote()), not just the fast path. This ensures the 3-second deadline covers the actual RPC calls that can hang indefinitely, even when the remote connection is already warm. 2. Fix available flag semantics: distinguish between "remote not reachable" (available=false) and "remote reachable but query failed" (available=true). A server that responds with a permission error or other non-branch-not-found error is still reachable and should not be reported as unavailable to users. Changes: - lore-transport: gRPC endpoint gets connect_timeout for initial connection - lore-revision: extract resolve_remote_latest() helper that returns (latest, authorized, available) where available reflects connection success rather than query success; eliminate double remote() call on event emission - tests: replace overfitted tautological test with two real unit tests Signed-off-by: Huân Lê-Vương <65440815+lehuan5062@users.noreply.github.com>
clippy::doc_markdown (denied via -D warnings in the pre-commit hook) flags bare type-name references in doc comments. CI caught this; local clippy runs missed it because they weren't re-run after this doc comment was added in a later revision of the same change. Signed-off-by: Huân Lê-Vương <65440815+lehuan5062@users.noreply.github.com>
Drop the 3-second status-level deadline from resolve_remote_latest and rely solely on transport-level connect timeouts (5s for gRPC handshake). The status-level timeout was layering two separate concerns and timing out slow lookups instead of bounding unreachable remotes. The root cause is QUIC's 30-second idle timeout on unreachable remotes: UDP offers no connection-refused signal, so every local status read would stall ~10–30s waiting for the timeout. Transport-level connect timeouts now bound this wait and are the single timeout layer. Update resolve_remote_latest to log previously-discarded errors at debug level before degrading gracefully. Remove the hung-remote timeout test (now invalid without a local timeout); add a failed-remote test mirroring terminal state from a bounded transport connect. Update doc comments to explain the degradation contract. Signed-off-by: Huân Lê-Vương <65440815+lehuan5062@users.noreply.github.com>
Lore's comment standard calls for minimal, purposeful doc comments and no restating of self-explanatory code. The rustdoc # Arguments/# Returns sections here described obvious parameters (repository, branch_id) and are rare elsewhere in the crate. Collapse to a terse summary that keeps only the non-obvious facts: graceful NoRemote degradation and the available flag reflecting connectivity rather than query success. Signed-off-by: Huân Lê-Vương <65440815+lehuan5062@users.noreply.github.com>
Several read verbs drove a remote connect even when local data is what gets returned, stalling for the bounded connect timeout against an unreachable server: - revision::resolve drove the connect before checking should_search_remote, so branch@LATEST under --local/--offline connected anyway; reorder the guard to match the @Number path. - Explicit-branch revision history connected unconditionally and then discarded the remote latest unless --remote was set; move the fetch inside the --remote branch. - File history connected before taking the --local early-return; compute the local latest and take the early-return first. - auth::resolve_user_info (driven by the CLI after history listings to prettify user ids) connected regardless of --local/--offline; skip resolution there and fall back to raw ids. With a killed server, revision history --local drops from ~4.1s to ~150ms; default-mode explicit-branch history no longer wastes a connect in the history lookup itself. Guarded by a new Python integration test against a killed server and a Rust unit test for the local-only branch@LATEST resolve. Signed-off-by: Huân Lê-Vương <65440815+lehuan5062@users.noreply.github.com>
ajcarberry
added
the
area:core
Collaborator
|
@lehuan5062 Could you update this to the latest main code and resolve the conflicts? |
…p-connect Signed-off-by: Huân Lê-Vương <65440815+lehuan5062@users.noreply.github.com> # Conflicts: # lore-revision/src/repository/status.rs # lore-revision/src/revision/history.rs # lore-transport/src/grpc/mod.rs # lore-transport/src/quic/client.rs
epic-lore-bot
Bot
added
imported
|
Imported as Lore CR-499. |
epic-lore-bot Bot
pushed a commit
that referenced
this pull request
Aug 31, 2026
**Problem** #107 bounded the remote-connect stall with 5s transport timeouts, but several read verbs still drive a remote connect even when local data is what gets returned. Against an unreachable server, every such read still pays the bounded ~5-10s wait for nothing: the fetched remote value is discarded, or the connect happens before the code even checks whether the caller asked for local-only data. This slows down `lore log <branch>`, the desktop app, and [my lore-web](https://github.com/lehuan5062/HUANREAL-lore-web) branch graph. Under `--local`/`--offline` the connect is also a contract violation, not just wasted time. **Change** - `revision::resolve` (revision.rs): the `branch@LATEST`/`HEAD` path drove the connect before checking `should_search_remote`, so `--local`/`--offline` connected anyway. Reordered to match the sibling revision-number path (`branch@<n>`), which already did this correctly. - Explicit-branch revision history (revision/history.rs): connected unconditionally, then discarded the remote latest unless `--remote` was set. The fetch now only happens inside the `--remote` branch. - File history (file/history.rs): connected before taking the `--local` early-return. The local latest and the early-return now come first; default-mode reconciliation is unchanged. - auth::resolve_user_info (auth/userinfo.rs): the CLI resolves user display names after every history listing, and this drove a connect regardless of `--local`/`--offline`. It now skips resolution in those modes; callers fall back to the raw ids. **Validation** cargo build, cargo clippy -p lore-revision, cargo test -p lore-revision -> all clean. New Rust unit test in lore-revision/tests/revision.rs for the local-only `branch@LATEST` resolve. New integration test scripts/test/test_local_reads_skip_connect.py: seeds and pushes a branch on a dedicated server, kills it, then asserts `--local` reads return the correct local latest well under the connect timeout, and the default-mode read doesn't stack connect waits. Existing suites test_branch.py, test_dirty.py, test_branch_switch_reconnect.py -> 89 passed. One failure, test_branch_merge_carries_dirty_through_clean_merge, is a pre-existing stack overflow already present on main, unrelated to this PR. Measured against a killed server: - revision history --branch <name> --local: ~4.1s -> ~150ms - file history --local: ~125ms - --remote still queries the server and fails fast when it's down Verified default behavior with the server up is unchanged. **AI Assistance Disclosure** I used an LLM (Claude Code) to help with implementation, code comments, and documentation on this change, and to help word things clearly as a non-native English speaker. I reviewed and tested all the code myself for its correctness, security, and license compliance. ``` Imported-PR: #115 Imported-From: 05d2cd4 Imported-Base: 046de55 Imported-Merge: bc580d3 Imported-Merge-Strategy: verbatim Imported-Merged-Paths: 0 Imported-Author: Huân Lê-Vương (lehuan5062) Signed-off-by: Huân Lê-Vương <65440815+lehuan5062@users.noreply.github.com> GH-URL: #115 ``` Lore-RevId: 850 Lore-Signature: 32ca666e6914bccad2995f2252f988a5df9f78ff5f446ba11cc3edca7c5a508f
Collaborator
|
Closed by extended change in 1ebed42 |
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
3 participants
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.
Problem
#107 bounded the remote-connect stall with 5s transport timeouts, but several read verbs still drive a remote connect even when local data is what gets returned. Against an unreachable server, every such read still pays the bounded ~5-10s wait for nothing: the fetched remote value is discarded, or the connect happens before the code even checks whether the caller asked for local-only data. This slows down
lore log <branch>, the desktop app, and my lore-web branch graph. Under--local/--offlinethe connect is also a contract violation, not just wasted time.Change
revision::resolve(revision.rs): thebranch@LATEST/HEADpath drove the connect before checkingshould_search_remote, so--local/--offlineconnected anyway. Reordered to match the sibling revision-number path (branch@<n>), which already did this correctly.--remotewas set. The fetch now only happens inside the--remotebranch.--localearly-return. The local latest and the early-return now come first; default-mode reconciliation is unchanged.--local/--offline. It now skips resolution in those modes; callers fall back to the raw ids.Validation
cargo build, cargo clippy -p lore-revision, cargo test -p lore-revision -> all clean.
New Rust unit test in lore-revision/tests/revision.rs for the local-only
branch@LATESTresolve.New integration test scripts/test/test_local_reads_skip_connect.py: seeds and pushes a branch on a dedicated server, kills it, then asserts
--localreads return the correct local latest well under the connect timeout, and the default-mode read doesn't stack connect waits.Existing suites test_branch.py, test_dirty.py, test_branch_switch_reconnect.py -> 89 passed. One failure, test_branch_merge_carries_dirty_through_clean_merge, is a pre-existing stack overflow already present on main, unrelated to this PR.
Measured against a killed server:
Verified default behavior with the server up is unchanged.
AI Assistance Disclosure
I used an LLM (Claude Code) to help with implementation, code comments, and documentation on this change, and to help word things clearly as a non-native English speaker. I reviewed and tested all the code myself for its correctness, security, and license compliance.