Consume the authenticated shared NODE stream for the owner-only globe pin, WatchWorkers for worker details, and the node snapshot for the public city estimate.
Remove repeated GetNodeLocation, GetNodeInfo, and ListWorkers requests on every node/peer snapshot; retain one-shot fallbacks only while a stream is unavailable.
Preserve stable globe state when peer updates do not change a location and document the stream/retry behavior.
Part of #741; this does not complete its broader multi-client acceptance criteria.
Verification
npm --prefix apps/sisyphus run typecheck
npm --prefix apps/sisyphus test
npm --prefix apps/sisyphus run build
make lint
actionlint
git diff --check
## Summary
- Consume the authenticated shared `NODE` stream for the owner-only globe pin, `WatchWorkers` for worker details, and the node snapshot for the public city estimate.
- Remove repeated `GetNodeLocation`, `GetNodeInfo`, and `ListWorkers` requests on every node/peer snapshot; retain one-shot fallbacks only while a stream is unavailable.
- Preserve stable globe state when peer updates do not change a location and document the stream/retry behavior.
Part of #741; this does not complete its broader multi-client acceptance criteria.
## Verification
- `npm --prefix apps/sisyphus run typecheck`
- `npm --prefix apps/sisyphus test`
- `npm --prefix apps/sisyphus run build`
- `make lint`
- `actionlint`
- `git diff --check`
Workspace globe data now follows the targeted live sources: the authenticated shared NODE stream for the owner pin, WatchWorkers for worker metadata, and the existing node snapshot for the public city estimate. This removes per-snapshot GetNodeInfo/GetNodeLocation/ListWorkers calls and keeps one-shot recovery only for stream outages. This is a focused follow-up under #741; the broader client-sync acceptance is still open. /cc @rvnminers-A-and-N
Workspace globe data now follows the targeted live sources: the authenticated shared `NODE` stream for the owner pin, `WatchWorkers` for worker metadata, and the existing node snapshot for the public city estimate. This removes per-snapshot `GetNodeInfo`/`GetNodeLocation`/`ListWorkers` calls and keeps one-shot recovery only for stream outages. This is a focused follow-up under #741; the broader client-sync acceptance is still open. /cc @rvnminers-A-and-N
Follow-up validation caught and fixed an additional conversion gap: browser NodeSnapshot mapping was dropping cityEstimate from both the initial GetNodeInfo response and WatchNodeState updates. That field now survives both conversion paths (and the dev proxy type), so the Workspace can actually consume it. Re-ran typecheck, all client tests, build, lint, actionlint and diff checks successfully.
Follow-up validation caught and fixed an additional conversion gap: browser `NodeSnapshot` mapping was dropping `cityEstimate` from both the initial GetNodeInfo response and WatchNodeState updates. That field now survives both conversion paths (and the dev proxy type), so the Workspace can actually consume it. Re-ran typecheck, all client tests, build, lint, actionlint and diff checks successfully.
Refreshed this PR onto current dev (including #748), resolving the shared-state enum overlap by preserving both NODE and MODEL_PROFILES. Validation on the refreshed branch passed: client typecheck, all 235 client tests, production build, make lint, actionlint, and git diff --check. Fresh CI is now running.
Refreshed this PR onto current `dev` (including #748), resolving the shared-state enum overlap by preserving both `NODE` and `MODEL_PROFILES`. Validation on the refreshed branch passed: client typecheck, all 235 client tests, production build, `make lint`, `actionlint`, and `git diff --check`. Fresh CI is now running.
/cc @rvnminers-A-and-N
Coordination note: #769 extends shared-state-domain.ts with the MODEL_CATALOG mapping (domain 16) on top of the #741 stream stack; this PR adds the NODE mapping (domain 1). There is no semantic collision, but the merged helper needs to retain both additions. I will leave this branch untouched and reconcile the two after the current CI/review sequence settles. /cc @rvnminers-A-and-N
Coordination note: #769 extends `shared-state-domain.ts` with the `MODEL_CATALOG` mapping (domain 16) on top of the #741 stream stack; this PR adds the `NODE` mapping (domain 1). There is no semantic collision, but the merged helper needs to retain both additions. I will leave this branch untouched and reconcile the two after the current CI/review sequence settles. /cc @rvnminers-A-and-N
Really like where this lands. There are no more GetNodeLocation/GetNodeInfo/ListWorkers calls on every peer tick. I checked that the NODE stream carries owner_location for the token holder and that pin edits fire locationChanged(), so the pin stays live.
Two things before it can go in:
It conflicts with dev again: #749 and #750 touched shared-state-domain.ts and its test.
In that same file, the case 1: case 'SHARED_STATE_DOMAIN_NODE': return 'node' block appears twice (lines 7-10). It looks like a leftover from the last conflict resolution. It's harmless, but the second copy is dead code and no-duplicate-case would flag it.
Since the whole #741 stack also edits that switch, do you want to land this first and rebase the stack on it, or fold the NODE mapping into the bottom of the stack? Also, on a node that isn't coordinating, does WatchWorkers end with an error? If so, the workspace would retry it every 30 s while visible. I wanted to check whether that's expected.
Really like where this lands. There are no more GetNodeLocation/GetNodeInfo/ListWorkers calls on every peer tick. I checked that the NODE stream carries `owner_location` for the token holder and that pin edits fire `locationChanged()`, so the pin stays live.
Two things before it can go in:
- It conflicts with `dev` again: #749 and #750 touched `shared-state-domain.ts` and its test.
- In that same file, the `case 1: case 'SHARED_STATE_DOMAIN_NODE': return 'node'` block appears twice (lines 7-10). It looks like a leftover from the last conflict resolution. It's harmless, but the second copy is dead code and `no-duplicate-case` would flag it.
Since the whole #741 stack also edits that switch, do you want to land this first and rebase the stack on it, or fold the NODE mapping into the bottom of the stack? Also, on a node that isn't coordinating, does `WatchWorkers` end with an error? If so, the workspace would retry it every 30 s while visible. I wanted to check whether that's expected.
I folded the workspace worker feed into the authenticated shared-state stream as well. When a node has no worker/pool service, WORKERS now reports one unavailable-domain event and keeps the stream open; a later successful pool join wakes it, instead of the page retrying WatchWorkers every backoff interval. Transport failures still use bounded reconnect, with a one-shot worker-list fallback only during the outage.
I also rebased this to a clean, linear branch on current dev. The shared-state domain mapper now has exactly one NODE case (it was duplicated in the old conflict resolution). The renderer receives the per-domain error without triggering a reconnect loop, so no-coordinator worker availability is expected and handled.
I folded the workspace worker feed into the authenticated shared-state stream as well. When a node has no worker/pool service, `WORKERS` now reports one unavailable-domain event and keeps the stream open; a later successful pool join wakes it, instead of the page retrying `WatchWorkers` every backoff interval. Transport failures still use bounded reconnect, with a one-shot worker-list fallback only during the outage.
I also rebased this to a clean, linear branch on current `dev`. The shared-state domain mapper now has exactly one NODE case (it was duplicated in the old conflict resolution). The renderer receives the per-domain error without triggering a reconnect loop, so no-coordinator worker availability is expected and handled.
/cc @rvnminers-A-and-N
Rebased the three workspace live-state commits onto current dev. The overlap with the merged shared-state enum is resolved while retaining both NODE and LOCATION_DISCOVERY; the duplicate NODE case is absent. The focused shared-state-domain test passes. Fresh CI is starting; the temporary checkout has no standalone dependency install, so I did not claim a full renderer test run. /cc @rvnminers-A-and-N
Rebased the three workspace live-state commits onto current `dev`. The overlap with the merged shared-state enum is resolved while retaining both `NODE` and `LOCATION_DISCOVERY`; the duplicate `NODE` case is absent. The focused shared-state-domain test passes. Fresh CI is starting; the temporary checkout has no standalone dependency install, so I did not claim a full renderer test run. /cc @rvnminers-A-and-N
Thanks for the careful review. I checked the current dev head (84ef65ec, including #749/#750): shared-state-domain.ts there has no NODE mapping, and this PR adds exactly one case 1 / SHARED_STATE_DOMAIN_NODE mapping. I also searched the PR diff and don't see a duplicate case; the earlier copy appears to have been removed during the restack.
On ordering, I would land this focused workspace PR directly on dev first, then restack the #741 stream stack on top. It is independently green and fixes the live workspace globe state; the stack is a separate series and should not need to be a prerequisite for it. I won't merge the daemon-backed stream stack before you review it.
WatchWorkers on a node without a coordinator ends with FailedPrecondition (apps/sisyphusd/api/local.go, with an explicit TestWatchWorkersRequiresAccessAndCoordinator). In this PR, that triggers one compatibility ListWorkers fallback and a bounded reconnect backoff (500 ms up to 30 s), not a fixed 30-second polling loop; once a stream update arrives the fallback is suppressed. The owner's pin has the same one-shot fallback behavior. The AccessPage draft dirty-state concern you raised on #753 is already addressed in the current PR diff by comparing effective draft values with the last saved config (accessDraftIsDirty), so reverting a toggle clears dirty state and remote updates replace a clean draft. #762 carries the separate coverage fixes you mentioned. Could you confirm you're comfortable with #761 landing first, and review the streaming stack when you have a chance?
Thanks for the careful review. I checked the current `dev` head (84ef65ec, including #749/#750): `shared-state-domain.ts` there has no `NODE` mapping, and this PR adds exactly one `case 1` / `SHARED_STATE_DOMAIN_NODE` mapping. I also searched the PR diff and don't see a duplicate case; the earlier copy appears to have been removed during the restack.
On ordering, I would land this focused workspace PR directly on `dev` first, then restack the #741 stream stack on top. It is independently green and fixes the live workspace globe state; the stack is a separate series and should not need to be a prerequisite for it. I won't merge the daemon-backed stream stack before you review it.
`WatchWorkers` on a node without a coordinator ends with `FailedPrecondition` (`apps/sisyphusd/api/local.go`, with an explicit `TestWatchWorkersRequiresAccessAndCoordinator`). In this PR, that triggers one compatibility `ListWorkers` fallback and a bounded reconnect backoff (500 ms up to 30 s), not a fixed 30-second polling loop; once a stream update arrives the fallback is suppressed. The owner's pin has the same one-shot fallback behavior. The AccessPage draft dirty-state concern you raised on #753 is already addressed in the current PR diff by comparing effective draft values with the last saved config (`accessDraftIsDirty`), so reverting a toggle clears dirty state and remote updates replace a clean draft. #762 carries the separate coverage fixes you mentioned. Could you confirm you're comfortable with #761 landing first, and review the streaming stack when you have a chance?
/cc @rvnminers-A-and-N
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
rvnminers-A-and-N
Summary
NODEstream for the owner-only globe pin,WatchWorkersfor worker details, and the node snapshot for the public city estimate.GetNodeLocation,GetNodeInfo, andListWorkersrequests on every node/peer snapshot; retain one-shot fallbacks only while a stream is unavailable.Part of #741; this does not complete its broader multi-client acceptance criteria.
Verification
npm --prefix apps/sisyphus run typechecknpm --prefix apps/sisyphus testnpm --prefix apps/sisyphus run buildmake lintactionlintgit diff --checkWorkspace globe data now follows the targeted live sources: the authenticated shared
NODEstream for the owner pin,WatchWorkersfor worker metadata, and the existing node snapshot for the public city estimate. This removes per-snapshotGetNodeInfo/GetNodeLocation/ListWorkerscalls and keeps one-shot recovery only for stream outages. This is a focused follow-up under #741; the broader client-sync acceptance is still open. /cc @rvnminers-A-and-NFollow-up validation caught and fixed an additional conversion gap: browser
NodeSnapshotmapping was droppingcityEstimatefrom both the initial GetNodeInfo response and WatchNodeState updates. That field now survives both conversion paths (and the dev proxy type), so the Workspace can actually consume it. Re-ran typecheck, all client tests, build, lint, actionlint and diff checks successfully.Refreshed this PR onto current
dev(including #748), resolving the shared-state enum overlap by preserving bothNODEandMODEL_PROFILES. Validation on the refreshed branch passed: client typecheck, all 235 client tests, production build,make lint,actionlint, andgit diff --check. Fresh CI is now running./cc @rvnminers-A-and-N
Coordination note: #769 extends
shared-state-domain.tswith theMODEL_CATALOGmapping (domain 16) on top of the #741 stream stack; this PR adds theNODEmapping (domain 1). There is no semantic collision, but the merged helper needs to retain both additions. I will leave this branch untouched and reconcile the two after the current CI/review sequence settles. /cc @rvnminers-A-and-NReally like where this lands. There are no more GetNodeLocation/GetNodeInfo/ListWorkers calls on every peer tick. I checked that the NODE stream carries
owner_locationfor the token holder and that pin edits firelocationChanged(), so the pin stays live.Two things before it can go in:
devagain: #749 and #750 touchedshared-state-domain.tsand its test.case 1: case 'SHARED_STATE_DOMAIN_NODE': return 'node'block appears twice (lines 7-10). It looks like a leftover from the last conflict resolution. It's harmless, but the second copy is dead code andno-duplicate-casewould flag it.Since the whole #741 stack also edits that switch, do you want to land this first and rebase the stack on it, or fold the NODE mapping into the bottom of the stack? Also, on a node that isn't coordinating, does
WatchWorkersend with an error? If so, the workspace would retry it every 30 s while visible. I wanted to check whether that's expected.I folded the workspace worker feed into the authenticated shared-state stream as well. When a node has no worker/pool service,
WORKERSnow reports one unavailable-domain event and keeps the stream open; a later successful pool join wakes it, instead of the page retryingWatchWorkersevery backoff interval. Transport failures still use bounded reconnect, with a one-shot worker-list fallback only during the outage.I also rebased this to a clean, linear branch on current
dev. The shared-state domain mapper now has exactly one NODE case (it was duplicated in the old conflict resolution). The renderer receives the per-domain error without triggering a reconnect loop, so no-coordinator worker availability is expected and handled./cc @rvnminers-A-and-N
Rebased the three workspace live-state commits onto current
dev. The overlap with the merged shared-state enum is resolved while retaining bothNODEandLOCATION_DISCOVERY; the duplicateNODEcase is absent. The focused shared-state-domain test passes. Fresh CI is starting; the temporary checkout has no standalone dependency install, so I did not claim a full renderer test run. /cc @rvnminers-A-and-NThanks for the careful review. I checked the current
devhead (84ef65ec, including #749/#750):shared-state-domain.tsthere has noNODEmapping, and this PR adds exactly onecase 1/SHARED_STATE_DOMAIN_NODEmapping. I also searched the PR diff and don't see a duplicate case; the earlier copy appears to have been removed during the restack.On ordering, I would land this focused workspace PR directly on
devfirst, then restack the #741 stream stack on top. It is independently green and fixes the live workspace globe state; the stack is a separate series and should not need to be a prerequisite for it. I won't merge the daemon-backed stream stack before you review it.WatchWorkerson a node without a coordinator ends withFailedPrecondition(apps/sisyphusd/api/local.go, with an explicitTestWatchWorkersRequiresAccessAndCoordinator). In this PR, that triggers one compatibilityListWorkersfallback and a bounded reconnect backoff (500 ms up to 30 s), not a fixed 30-second polling loop; once a stream update arrives the fallback is suppressed. The owner's pin has the same one-shot fallback behavior. The AccessPage draft dirty-state concern you raised on #753 is already addressed in the current PR diff by comparing effective draft values with the last saved config (accessDraftIsDirty), so reverting a toggle clears dirty state and remote updates replace a clean draft. #762 carries the separate coverage fixes you mentioned. Could you confirm you're comfortable with #761 landing first, and review the streaming stack when you have a chance?/cc @rvnminers-A-and-N