Repository navigation
feat: the umbrella's worker serves the default network zone; pin the zone releases - #27
Merged
leandrolerena merged 32 commits intoNov 7, 2026
Merged
Conversation
…mp the umbrella to 0.3.0 Network zones need all four together: core stores the zones, the engine routes jobs by zone, the monitor ships with the engine, and worker 0.4.0 reads NEOPS_WORKER_NETWORK_ZONES. These charts appear on Quay with the core v2.4.0, engine v0.45.0 and SDK v0.4.0 tags.
The central worker lists default in NEOPS_WORKER_NETWORK_ZONES in the umbrella's values, never in the neops-worker chart's defaults, so a per-zone release that misses its own setting is refused instead of serving default. The README documents one neops-worker release per zone, two workers per zone, escaping commas in --set, and creating a zone before its workers start.
Its Network zones page is where the README tells operators to create a zone. Like the other pins in the umbrella's 0.3.0, the chart appears on Quay only with its tag, so this waits on web client v6.1.0 next to core v2.4.0, engine v0.45.0 and SDK v0.4.0.
Web client 6.1.0 ships the page, so it leads the options; the mutation and Django admin remain for those without it.
Devices leave the zone while its workers still run, or by an override step or Django admin once the site is gone; then the zone's worker release goes, and the zone last. Otherwise the move waits up to an hour for a worker that never comes.
It admits at most 2 online workers, which the default zone alone fills.
This was referenced Oct 31, 2026
The chart defaults to quay.io/zebbra/neops-worker-sdk at its appVersion, which a KIND loop built from the working tree does not have; the umbrella's KIND values use neops-worker-sdk:cutting-edge.
From core 2.4.0, superuser.enabled also seeds the admin role, the workflow admin profile and write on network-zones, as compose's cms-init does plus the zone grant. Say so, and warn against starting a site's worker with default when nobody can create its zone.
…ies on every upgrade
…45.0 Engine 0.45.0 connects as POSTGRES_USER, which this chart sets to neops, while existing installs' tables belong to postgres. Without the hand-over the engine fails at startup with relation "mikro_orm_migrations" already exists. The README gives the kubectl exec form that runs the engine's owner-to-postgres-user.sql in the chart's Postgres pod, and the password change. The same commit says how the KIND zone release takes the two image flags.
Engine 0.44 always connects as postgres, so a rollback needs the engine's POSTGRES_PASSWORD set back to it, after the engine's own down migration. The tables handed to neops need nothing, as postgres is a superuser.
…d, and take the script from docs.neops.io The engine repository is private, so the upgrade script and the engine docs are linked on docs.neops.io.
Engine 0.45.0 and SDK 0.4.0 are released from develop without network zones, so the charts that carry zones come with the engine v0.46.0 and SDK v0.5.0 tags.
The POSTGRES_USER fix first ships in engine 0.45.0, not in the beta tagged before it merged.
…ations The zones migrations now ship in 0.46, and 0.45 adds one migration of its own, so a rollback to main's engine pin reverts both, newest first.
An install that jumped from 0.44.1 to 0.46 ran the rename after the zones migrations, so reverting those from the 0.46 image already undid it; a second down from the 0.45 image would target a migration no longer executed.
The down reverts every executed migration whose name sorts at or after its target, so 0.45's rename goes with the zones migrations whatever the install's history.
An engine without zones ignores a worker's zones, so a zone's worker would take any zone's jobs, and a 0.5 worker cannot read 0.44's jobs. Going forward again, the engine counts workers that registered before it in default until they restart.
The engine is stopped for the down, so there is no pod to exec into; the pod takes the 0.46 image, the Quay pull secret, the engine's database settings and its Secret.
…e the engine to 0.45.0 or later main pins core 2.3.0, which cannot run on 2.4.0's migrations, so the way back drains the log outbox and migrates core to 2.3.0's last migration first. The chart now jumps from engine 0.44.1 to 0.46.0, past 0.45.0.
Engine develop's refused-worker records (zebbra/neops-workflow-engine#507) add Migration20261102113953; it sorts after the zones migrations too, so the down reverts it and deletes the refused-worker rows.
2 of 4 tasks
…44.1 Give the engine page's SQL a command, and say to restore the neops password when going forward again.
Once this merges, main pins the zones release, so upgrading from main re-installs it.
The umbrella pairs engine 0.46 with worker 0.5; the gate compares the client the worker pins, not the worker's own version.
… migrations rollout status reports a Deployment scaled to 0 as done while its pod still terminates, so the SQL and the down could overlap a live engine.
Undoing 0206_network_zone deletes every zone and device zone, so zone releases scaled back up are refused until the zones exist again. Link core's 2.4.0 notes, which hold the rule, and recreate the zones and move the devices back first.
…uthorizes nothing The engine checks a worker's zones only at registration, so a running worker that lists none keeps claiming default jobs. Any pod or host that reaches the worker routes can register for any zone, so restrict and encrypt the path. Link the engine's Network Zones page for the rules.
core 2.4.0's bootstrap_admin_role gives the role it seeds the network-zones write also when other users hold it (zebbra/neops-core#2474); the hook still skips the workflow profile on its exit 3.
Every holder of the shared role, so under OIDC every user the identity provider puts in it, may write every zone; the seed re-applies the write on every upgrade.
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.
Description
v2.4.0, enginev0.46.0, SDKv0.5.0and web clientv6.1.0; each tag pushes its chart tooci://quay.io/zebbra/chartsmainNEOPS_WORKER_NETWORK_ZONES: defaultfor its central worker (charts/neops/values.yaml:115), never the neops-worker chart's defaults, so a per-zone release that misses its own setting is refused instead of servingdefault.charts/neops/Chart.yamlpins core 2.4.0, engine and monitor 0.46.0, worker 0.5.0 and web client 6.1.0, and bumps the umbrella to 0.3.0.network-zones(core 05 zebbra/neops-core#2469), re-applied on every install and upgrade; other users neednetwork-zoneswrite in a role of their own; and never start a site's worker withdefaultto work around a missing grant.kubectl runpod on the 0.46 image (the README gives the command), whose down to the first zones migration also reverts 0.45'sMigration20261102104651and, since enginedevelop's zebbra/neops-workflow-engine#507,Migration20261102113953, whose revert deletes the rows of refused workers (the revert runs in name order, and 0.44.1 applies neither again); set the engine's password back topostgres's (and toneops's again going forward); from the still-running 2.4.0 core, drain the log outbox andmigrate core 0204_merge_20261029_logging_develop, sincemainpins core 2.3.0; upgrade withmain's chart, so the central worker goes back to 0.3.0 with engine 0.44.1. Going forward again, restart the central worker and scale the zone releases back up, so every worker re-registers its zones.--set, creating a zone on the web client's Network zones page before its workers start, and the order for retiring a site.Look at first
charts/neops/values.yaml:115: the setting lives in the umbrella's values only.charts/neops/Chart.lockis gitignored, so the PR carries no lock. Once the charts are on Quay,helm dependency update charts/neopsresolves the new pins locally.Deviations from the design
None.
Testing
Manual lab integration run
zones-it, twice on 2026-10-31:75cf9bf57, engine029e791e, SDK725b28b, weba26da57;839a61400, engine50dc36d9, SDKcc98ac9, webebf30ad, lab2636d6a. The final heads differ only by one engine code comment and doc line, and the SDK's unit-tested formatter for list messages.No lab round exercised this PR: helm was not part of them. Live on KIND (namespace
neops-zones-test, since deleted), with the charts packaged locally over zones images: the umbrella installed; core's hook seeded the superuser's role, and the engine answered/workersand/network-zonesfor it; the README's zone-b release with--set nodeSelector.network-zone=zone-bran two workers on the labelled node, both online in zone-b; a selector naming a label no node has left its pod Pending. Not verified:helm dependency updateand the README's Quay-chart form, which need the charts on Quay; the published images. Since the KIND control-plane node is tainted, the positive scheduling alone doesn't prove the selector; the Pending case does. The README's "Going back to engine 0.44.1" recipe was not rehearsed on a cluster: neither the one-offkubectl runengine down normigrate core 0204_merge_20261029_logging_developfrom the 2.4.0 pod. The lab proved the engine down under compose only.Network Zones train, in merge order
v0.45.0fromdevelopfix/pin-engine-client-0.45.0f7cdf41, client==0.45.0); SDKv0.4.0is tagged on it and released (2026-11-06); Claude has merged it into SDK 01–03fix/pyasn1-0.6.5feature/network-zones-01-modelfeature/network-zones-02-run-writesfeature/network-zones-03-search-and-lockfeature/network-zones-04-zone-paging-and-sortfeature/network-zones-05-chart-admin-rolev2.4.0-beta.1is tagged on its tip (d2b030b19)feature/network-zones-06-unlock-write-refusedv2.4.0-beta.2is tagged on its tip (da4974bb6): the unlock answersWRITE_REFUSED, which engine 04 needsfeature/network-zones-01-nest-client-2.3.0fix/postgres-userv0.45.0-beta.1;v0.45.0is the first release that holds itfeature/network-zones-02-definitionsfeature/network-zones-03-routingfeature/network-zones-04-device-writesfeature/network-zones-05-prerelease-versionsfeature/network-zones-06-refuse-invalid-pushesfeature/network-zones-07-cms-request-timeoutfeature/network-zones-08-never-polled-and-log-contextfeature/network-zones-09-monitor-contextfeature/network-zones-10-cms-at-lock-and-write-backv0.46.0-beta.3is tagged on its tip (4c4f2f7c) and published client 0.46.0b3; the engine accepts the SDK's 0.46.0b2feature/network-zones-01-job-metav0.4.0is released; the user merges 01–03 back to backfeature/network-zones-02-zonesfeature/network-zones-03-strict-compatibilityv0.5.0-beta.2is tagged on its tip (907d443), on client 0.46.0b2feature/network-zones-01-angular-client-2.3.0feature/network-zones-02-zonesv6.1.0-beta.1is tagged on its tip (953db30)feature/network-zones-01-sdk-0.4developpins SDK==0.4.0b1fix/network-zones-sdk-0.5.0b1-pin==0.5.0b2,uv lock; the user merges itv2.4.0MINIMUM_CORE_VERSION2.4.0, generated pagednetworkZones, then tag enginev0.46.0==0.46.0, then tag SDKv0.5.0==0.5.02.4.0exactly, library peer floor back to>=2.4.0 <3, plainnpm install, re-renderdocs/dashboard-cards/table.mdagainst corev2.4.0, then tag webv6.1.0feature/network-zones-01-scenariofeature/network-zones-01-central-worker-zonefeature/network-zones-02-pinsfeature/network-zones-01-central-worker-zonePhase A (steps 1–5) ships the zones betas: Claude tags each on the tip of its stack once CI is green, and nothing merges into
developfor a beta. Claude keeps every zones branch current by mergingdevelopin. Merges intodevelopormain, step 0 (closing the 0.45 line fromdevelopbefore engine 02 or SDK 01 merges) and every final tag are the user's. Phase B (6–10) ships the finals, and Phase C (11) the deployments that pin them.Checklist: