feat: tell which members a GroupV2 run left on which branch - #223
Draft
osmaczko wants to merge 2 commits into
Draft
feat: tell which members a GroupV2 run left on which branch#223osmaczko wants to merge 2 commits into
osmaczko wants to merge 2 commits into
Conversation
osmaczko
force-pushed
the
feat/groupv2-fork-report
branch
from
August 24, 2026 15:57
1125c79 to
23feb28
Compare
A GroupV2 group can divide into branches that never converge again, and no member can tell from its own view that it happened: both halves keep a roster and an epoch number, each simply stops being able to read what the other sends, and a fresh split leaves both carrying the same members. A member that has merely fallen behind produces the same endless run of undecryptable frames, from a different cause needing a different fix, and nothing separates the two. MLS derives a value for exactly this comparison, meant to be compared out of band: at one epoch number, different epoch authenticators are a fork and identical ones are real convergence. Write it once per epoch a conversation reaches, so a run leaves behind the one thing that partitions its members.
The epoch authenticator answers the question one member at a time, and the question is about the group: how many branches a run ended in, when each split off, and who is on it. Answering it means putting every member's log side by side, which stops being something to do by hand as soon as a group is more than a handful. Read the epoch lines out of a run, group members by the authenticator they held at each epoch, and render the result as a lineage tree and one lane per member. Distinguish a member that forked from one that only fell behind, since both drop every frame the group sends and the drop count cannot tell them apart. Take either one log per member or a single log split on its user_id spans, so the same report runs on the in-process group tests as on a farm of real peers.
osmaczko
force-pushed
the
feat/groupv2-fork-report
branch
from
August 24, 2026 16:23
23feb28 to
e9463be
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.
A GroupV2 group can divide into branches that never converge again, and no member can tell from its own view that it happened: both halves keep a roster and an epoch number, each simply stops being able to read what the other sends, and a fresh split leaves both carrying the same members. A member that has merely fallen behind produces the same endless run of undecryptable frames, from a different cause needing a different fix, and nothing separates the two.
MLS derives a value for exactly this comparison, meant to be compared out of band: at one epoch number, different epoch authenticators are a fork and identical ones are real convergence. Write it once per epoch a conversation reaches, so a run leaves behind the one thing that partitions its members.
The value answers the question one member at a time, and the question is about the group: how many branches a run ended in, when each split off, and who is on it. Answering it means putting every member's log side by side, which stops being something to do by hand as soon as a group is more than a handful.
tools/forks.pyreads the epoch lines out of a run, groups members by the authenticator they held at each epoch, and renders the result as a lineage tree and one lane per member, distinguishing a member that forked from one that only fell behind. It takes either one log per member or a single log split on itsuser_idspans, so the same report runs on the in-process group tests as on a farm of real peers.Note for the review, not for history: this is based on
feat/surface-group-progress-events(#219), which is not merged. Retarget once that lands.