test(probe): drive several conversations from one process on both ends - #38
Merged
Conversation
`dial` opens one conversation, which is the one shape that cannot show the failure the shared node was built to fix: the app runs several at once from a single process under a single identity. `dial-many` accepts N invites in one runtime, requires every session to reach ready and every message to be delivered, and reports the listen port each session's node bound. That last part is the measurement. Run against the same host, same script: v0.7.2 distinct_nodes: 3, ports [62028, 62035, 62043] v0.7.3 distinct_nodes: 1, ports [62482, 62482, 62482] The runner grows `--sessions N`, which starts N independent remote `listen` processes — each its own process and so its own counterpart — and dials them all from one local process. Known limit, and the reason this is a probe rather than a gate: with every counterpart on ONE host, three concurrent conversations do not converge, on v0.7.2 (0 of 3 ready) any more than on v0.7.3. Colocated peers behind a single address are a configuration the substrate already handles badly, so this harness can prove the topology and single-conversation delivery, not multi-conversation delivery. Counterparts on distinct hosts would close that.
The far end was N separate `listen` processes on one host, which is N nodes behind one address — the exact shape this work removed. A test whose far end reproduces the defect cannot measure the fix, and it did not: three conversations failed to converge on v0.7.3 and on v0.7.2 alike, which I read as "the harness cannot show this" when it actually said "the harness is broken". `listen-many` serves N invites from one runtime, so both ends are now one node with N rooms. Same script, same host: 3 sessions local 1 node (52935 x3), remote 1 node (52373 x3) delivered 6 sessions local 1 node (52363 x6), remote 1 node (37422 x6) delivered and on v0.7.2, three sessions are three nodes (62028, 62035, 62043) with 0 of 3 ready. Both ends print their topology, so a run that quietly falls back to a node per session is visible rather than inferred.
ForeverInLaw
changed the title
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
1 participant
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
dialopens one conversation — the one shape that cannot show the failure the shared node was built to fix. The app runs several at once from a single process under a single identity.dial-manyaccepts N invites in one runtime and requires every session ready and every message delivered.listen-manyserves N invites from one runtime, so the far end is one node too.--sessions Nand prints both ends' topology.The correction this PR went through
The first version ran N separate
listenprocesses on one host. That is N nodes behind one address — the exact shape this work removed — so the far end was reproducing the defect the run was trying to measure. Three conversations failed to converge on v0.7.3 and on v0.7.2 alike, which I first read as "the harness cannot show this". It actually meant "the harness is broken". Fixed by giving the remote one process for all sessions.What it measures now
Same script, same host, one node per side:
Against v0.7.2, built from the tag with the same harness: 3 nodes (62028, 62035, 62043), 0 of 3 ready.
Both ends print their topology, so a run that quietly falls back to a node per session is visible rather than inferred.
Note
Deploying this to a probe host needs the matching
libmoss.so(>= v0.8.19) beside the binary. A stale one fails cleanly withMoss symbol unavailable: Moss_JoinRoom— which is how the first server run failed, and the guard working as intended.