feat: raise MaxFrameSize 256 MiB → 1 GiB + env override#22
Open
TeoSlayer wants to merge 2 commits into
Open
Conversation
Bumps the per-frame cap from 256 MiB → 1 GiB (default) so single-frame send-file works for the larger artefacts operators are actually shipping on multi-GiB-RAM hosts. Adds PILOT_DATAEXCHANGE_MAX_FRAME env override (read at package init, range-checked to [64 KiB, 2 GiB)) so the limit can be raised further without recompiling. Wire format is unchanged — the length field has always been uint32, so the absolute ceiling has always been ~4 GiB; this cap is a memory-safety knob, not a wire-format limit. Both ends must agree: a sender that exceeds the receiver's cap still gets "frame too large" and dropped. The chunked streaming protocol in docs/PROPOSAL-reliable-file-transfer.md (web4) will eventually remove the need for a per-frame cap entirely. Tests: - TestReadFrame_SizeCap updated for the new default (was 256 MiB). - New TestMaxFrameSize_DefaultAndConstants pins the default and verifies the env override is unset in CI. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Codecov Report✅ All modified and coverable lines are covered by tests. 📢 Thoughts on this report? Let us know! |
Adds TypeFileStream (frame type 7) — a chunked transfer protocol that replaces the single giant TypeFile frame that stalled on any non-trivial path (>64 KiB stalls over relay, and over a direct link that flips to relay under one-way load). - filestream.go: INIT/CHUNK/ACK/DONE/COMPLETE control protocol; sender state machine (SendFileStream) with a sliding window, end-to-end SHA-256, and resume keyed on a content-derived transfer_id; receiver (StreamReceiver) writing .partial contiguously, verifying the full hash, and atomically renaming into place. Chunk size 48 KiB — deliberately below the ~64 KiB tunnel-stall threshold, so each chunk rides the known-good path and the per-chunk ACK keeps the reverse direction busy (which also stops the blackhole heuristic flipping the link mid-transfer). - service.go: dispatch TypeFileStream in handleConn via a per-connection StreamReceiver; final filename + file.received event match the legacy saveReceivedFile path. - zz_filestream_test.go: loopback tests — happy path, resume-from-offset, corrupt-resume rejection, codec round-trip. Verified on the Mac↔GCP-VM dual-NAT rig: 50 MB completes with matching sha256 over a hole-punched direct path; resume from a mid-transfer kill continues from the exact byte offset. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
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
2 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.
What
Why
The wire format's length field has always been `uint32` — the absolute ceiling is ~4 GiB. The 256 MiB cap was a memory-safety knob sized for "the test fleet's 100 MiB file payloads" (per the old comment). Operators on multi-GiB-RAM hosts are hitting the cap on real workloads (the "send-file fails on 50 MB / 10 MB files" report in `web4/docs/BUG-updater-version-skew.md` is a separate reliability issue, but raising the cap removes one ceiling on the same code path).
The chunked streaming protocol in `web4/docs/PROPOSAL-reliable-file-transfer.md` will eventually remove the need for a per-frame cap entirely — this is the smallest change that unblocks bigger transfers today without changing wire format.
Compatibility
Wire format unchanged. Both ends must agree on the cap:
Tests
🤖 Generated with Claude Code