What it means
Every server restart is a new stream of the journal, so runtimes compact away streams that have been dead for long. Compaction takes away the records of those streams and nothing anyone still relies on: a claim that holds and a session that is open are never touched; a handover nobody has taken keeps its stream for a day past the retention, and no longer, so a machine whose runs each handed over something cannot fill its quota with the dead; a handover still offered keeps everyone who took it, and a review keeps every answer, even when the taker's or the answerer's runtime has stopped. And a runtime compacted away while it slept — a laptop past the expiry and the fifteen-minute retention — is heard again when it wakes: its peers fetch its stream again from the first event, so a claim it took before it slept holds again there and refuses a conflicting one.
How it works
Journal::compact (apps/majordomus-cli/src/mesh/journal.rs) considers only streams silent
for longer than the expiry plus the retention. Of those, it keeps a stream that offers a
handover the fold lists with no taker, while its silence is within HANDOVER_RETENTION past
the retention, and — to a fixed point — a stream holding the taking of a handover or the
answer to a review whose own stream is kept. Who took and who answered is read by
mesh::state::fold, the reading every surface shows. What goes leaves a tombstone: the
sequence it reached, advertised so it is not sent again, and the beat it was last heard at.
A late event of a compacted stream is absorbed into the tombstone. Journal::merge_marks
lifts a tombstone on a verified, trusted beat that is fresh and above the tombstone's beat;
the stream's mark falls to nothing held and the peers send it again from its first event.
The server's supervisor compacts every sixty heartbeats with PEER_RETENTION and
HANDOVER_RETENTION (apps/majordomus-cli/src/mesh/cooperation.rs).
How to see it
test/run.sh 493_mesh_compaction_keeps_what_is_live
cargo test --manifest-path apps/majordomus-cli/Cargo.toml --test mesh_compaction
cargo test --manifest-path apps/majordomus-cli/Cargo.toml --lib mesh::journalThe case runs tests/mesh_compaction.rs: two runtimes linked through the real signed
protocol over an in-process transport, one compacted while it sleeps and heard again whole
when it wakes, a taker and a reviewer that stop while the publisher runs, and two stopped
publishers of which only the one whose handover nobody took is kept. The journal's unit
tests reproduce each defect PR 588 recorded, and its property test,
compacting_a_crowded_node_changes_nothing_live, holds over random histories that no live
claim, open session, handover taker or review answer changes.
What it does not cover
No black-box case reaches compaction in bounded time — a server compacts every sixty heartbeats and keeps a dead stream for fifteen minutes, and only the heartbeat and the expiry are declarable — so the proof is the library's own runtimes rather than two servers. Two runtimes can print different digests while each compacts on its own schedule: one has forgotten a dead stream's ended records that the other still lists. What is live agrees, and marks agree. A review request of a dead runtime keeps nothing on its own and goes with its stream. Tombstones are not persisted: a restarted runtime may be sent a compacted stream again and drops it again later.
Why it exists
Compaction is what keeps a machine that restarts its servers from filling its quota, and a bound that forgets what is live is worse than no bound: a handover that looks untaken is taken twice, a review that opens again is answered twice, and a runtime that wakes up unheard holds claims its peers cannot see. PR 588 recorded all three; this claim is their closure.