The errors league (GlitchTip) and the uptime/status league
Status: RULED AND EXECUTED 2026-08-11. The body below is the 2026-08-06 worksheet, preserved as written. What it proposed has since been ruled by the owner and carried out; the execution record is §0.
Date: 2026-08-06. Measured against shc origin/main @ de52c6a24. All
upstream facts were verified this session against GlitchTip primary sources
(URLs + access date inline); the split between verified and assumed is §7.
0. EXECUTION RECORD: 2026-08-11
Section titled “0. EXECUTION RECORD: 2026-08-11”The owner ruled §1 as written and additionally ruled the §4 sub-question conditionally: GlitchTip takes the uptime/status league too if its public status page is real and good; otherwise the leagues split. The condition was re-measured before deciding, because §2.3 measured v6.2.3 and the package has since moved to v6.2.5.
Re-measurement, 2026-08-11, at the pinned v6.2.5 and re-confirmed at v6.2.6 (the newest tag at the time):
| Probe | Result |
|---|---|
glitchtip/settings.py (master) searched for status-page / public / anonymous settings | none. What exists is GLITCHTIP_ENABLE_UPTIME (True), UPTIME_CHECK_INTERVAL, GLITCHTIP_UPTIME_RETENTION_DAYS, and the two SSRF flags, all about checking other endpoints, none about publishing this instance’s state |
| upstream issue #36 “Public Status Page” | still OPEN, label Help Wanted, filed 2022-08-26, unimplemented; the reporter’s own last word is that they went with another product |
| release blog, 5.0 (2025-05) → 6.2 (2026-06) | no release note mentions a status page |
tag list on glitchtip/glitchtip-backend | v6.2.5 (2026-08-07) is the pin; v6.2.6 (2026-08-07) is newest; the line is alive, one maintainer |
So the condition FAILED and the leagues SPLIT, exactly as §4 recommended. GlitchTip has genuine internal uptime monitoring and no anonymous surface of any kind; there is nothing upstream for a package to expose, so this is not a packaging gap that could ever be closed here.
What was applied:
flagship: errors→apps/monitoring/glitchtip/app.yaml. Access posture unchanged and re-affirmed:access: open, errors stay PRIVATE behind GlitchTip’s own auth, because a forward-auth gate would 401 every SDK on the DSN ingestion endpoints. The measurement above is transcribed into that app.yaml, pinned to the version, so a future bump that adds a status page has to come back through it and move the league.flagship: uptime→apps/monitoring/uptime-kuma/app.yaml, together with the two things §4’s “owed” paragraph named: a top-levelaccess: gatedand the# flagship-userless:marker the fitness check demands in addition to it.- And the thing §4 did not name, which the label made mandatory:
uptime-kuma’s public status page was NOT actually public. ce’s disposition
default is already
gated, so every path including/status/<slug>sat behind the forward-auth gate; the league would have been awarded for a capability the packaging withheld.ungated_pathsnow carves out exactly the anonymous read surface (/status/,/api/status-page/,/api/badge/,/assets/,/upload/), read off the pinned 2.5.0 source;/socket.io/(every write in the product),/metricsand/api/push/stay gated. See the app.yaml block for the per-prefix argument. highlightRETIRED: deleted from the catalogue with its line inapps/RETIRED. It is installed on NEITHER estate (both ops repos grepped), so nothing needed migrating.
Not changed, deliberately: no flagship: key was added to any app this doc
does not name, and the operator advice in §4 stands: run both. GlitchTip is
the private, OIDC-gated error tracker; uptime-kuma is the public status page.
1. The verdict, up front
Section titled “1. The verdict, up front”errors→ GlitchTip. GlitchTip clears every mechanical flagship bar and, crucially, clears the bar highlight FAILED: its SSO is MIT and unpaid. It is the strongest errors candidate the catalogue has ever carried; the zero-touch OIDC path is authored in-package (a DB-seed job, no click-ops), not merely aspired to. The one owed item is the universal one: a from-empty install+login proven live, which no candidate in the flagship table has today.uptime/status→ SPLIT: keep uptime-kuma. GlitchTip has genuine uptime MONITORING (Ping/GET/POST/Heartbeat + alerting) but no public status page, the shareable, anonymous, read-only surface that defines a status league. That is not a packaging gap; the feature does not exist upstream (GitLab issue #36, open since 2022). uptime-kuma ships first-class public status pages and a landedauth.disable. So GlitchTip should NOT take both leagues:errors= GlitchTip,uptime= uptime-kuma. GlitchTip’s own uptime checks remain a useful feature OF the errors flagship (co-located alerting), not a claim on the status league.
2. GlitchTip: the evidence
Section titled “2. GlitchTip: the evidence”2.1 License: MIT, and the “license key” gates nothing
Section titled “2.1 License: MIT, and the “license key” gates nothing”- MIT, verbatim
MIT License / Copyright (c) 2019 GlitchTip(https://gitlab.com/glitchtip/glitchtip-backend/-/raw/master/LICENSE, 2026-08-06). Modification + redistribution + self-hosting permitted; clears flagship requirement 4. - The license key is voluntary support, not a feature gate.
settings.py(fetched from-/raw/master/glitchtip/settings.py, 2026-08-06) comments theGLITCHTIP_LICENSE_KEYblock: “Hides the ‘Support GlitchTip’ banner and enables future support features.” The install docs on the legacy aliasI_PAID_FOR_GLITCHTIPsay plainly: “This won’t enable extra features.” (https://glitchtip.com/documentation/install). The 6.1 blog (https://glitchtip.com/blog/2026-03-23-glitchtip-6-1-released/) frames it as a support-prioritisation ID, and the support page (https://glitchtip.com/support/) only requires it to use the support form.BILLING_ENABLEDisFalseuntil Stripe keys are supplied (settings.py), so a self-host with no keys is fully functional. Nothing, including SSO, is behind a paid tier. This is exactly the bar highlight failed (flagship-tier.md§7: “highlight FAILS the no-paid-SSO bar”).
2.2 SSO: zero-touch OIDC in the free self-host, no click-ops
Section titled “2.2 SSO: zero-touch OIDC in the free self-host, no click-ops”- GlitchTip is
django-allauth(allauth.socialaccount.providers.openid_connectis inINSTALLED_APPS,settings.py:616). SSO is the same MIT code for everyone, not paywalled. - But it is NOT env-drivable in the stock image.
settings.py:1226-1234buildsSOCIALACCOUNT_PROVIDERSfrom only four hard-coded env hooks (gitlab, gitea, nextcloud, microsoft); there is no genericopenid_connect/APPSenv path and noSOCIALACCOUNT_PROVIDERSJSON env. The documented route is the Django admin at/admin/socialaccount/socialapp/, entering a{"server_url": "…/.well-known/openid-configuration"}settings blob (https://glitchtip.com/documentation/install), i.e. click-ops. - So the zero-touch route is to seed the
socialaccount_socialapprow directly, the exact object that admin form writes.apps/glitchtip/oidc.plug.yamldoes this via amanage.py shelljob (context: service, idempotentupdate_or_createkeyed onprovider='openid_connect', provider_id='keycloak'). This is safe and restart-free because allauth’s OIDC callback is a per-request parameterised route (^(?P<provider_id>[^/]+)/…/login/callback/, verified against allauth’sopenid_connect/urls.pyanddocs.allauth.org, 2026-08-06); the seeded row is resolved on the next request, no server bounce. GlitchTip uses nodjango.contrib.sites, so the row needs no site M2M. The registered redirect URI is<origin>/accounts/oidc/keycloak/login/callback/(the plug publishes exactly that to ce’screate-oidc-client). Result:shc install glitchtip -i oidc=<keycloak>wires working Keycloak SSO with zero human steps, a stronger zero-touch story than any candidate in the flagship table, which all carry an “owed: zero-touch proof” note.
2.3 Uptime monitoring: yes. Public status page: NO.
Section titled “2.3 Uptime monitoring: yes. Public status page: NO.”- Uptime monitoring is first-class.
apps.uptimeis enabled by default (GLITCHTIP_ENABLE_UPTIME=True,settings.py:267); monitor types are Ping (HTTP HEAD), GET, POST and Heartbeat, with email/webhook alerting (https://glitchtip.com/documentation/uptime-monitoring/, 2026-08-06). Monitors are project-scoped and entirely auth-gated. - There is no public status page. Verified two ways: (a) the uptime docs
describe no public/anonymous/shareable status surface; (b) GitLab issue #36
“Public Status Page” is still OPEN, label Help Wanted, created
2022-08-26 (
https://gitlab.com/glitchtip/glitchtip/-/issues/36). The request text asks for exactly the missing capability: “Ability to create an uptime state page and mark which projects/individual monitors show on it.” - The privacy-model consequence. The question was: can an uptime surface be PUBLIC while error projects stay PRIVATE? For GlitchTip the answer is moot: there is no anonymous surface at all. Everything (errors and uptime) sits behind login inside the Organization → Team → Project hierarchy; there is no “public” flag on a monitor or a status page. So GlitchTip cannot offer a public status page without also… it simply cannot offer one, period.
2.4 Maintenance health: actively maintained
Section titled “2.4 Maintenance health: actively maintained”- Latest release v6.2.3, tagged 2026-07-31; the line has advanced steadily
(6.1.0 2026-03-21 → 6.2.0 2026-06-19 → 6.2.3), with commits landing on
2026-08-06 (release blog
https://glitchtip.com/blog/; GitLabglitchtip/glitchtip-backendactivity). Sole active author is the founder/maintainer David Burke; bus-factor is effectively one, an honest note, but the project is unambiguously alive and not a-classicline. This is the direct contrast with highlight’s post-acquisition self-host drift.
3. GlitchTip against the eight flagship bars (flagship-tier.md §1)
Section titled “3. GlitchTip against the eight flagship bars (flagship-tier.md §1)”| Bar | Assessment |
|---|---|
| 1 SSO ladder, unpaid | PASS, best-in-class. MIT OIDC, unpaid (§2.1, §2.2). Native login form + Keycloak via the oidc plug. No auto-redirect knob upstream (login is an allauth/headless SPA), so the rung is “visible SSO button”, not auto-redirect, but reachable without a license, which is the bar highlight failed. |
| 2 Zero-touch: the hard bar | Mechanism PASS; live proof OWED. The DB-seed job wires SSO from -i oidc=<keycloak> with no click-ops (§2.2), and glitchtip-init seeds a break-glass superuser so the deploy is usable from empty. Not yet proven from-empty with a live browser login (the universal gap; no candidate has this today). |
| 3 One app per league | PASS. errors is vacant; no other claimant. |
| 4 License | PASS. MIT (§2.1). |
| 5 Lifecycle honesty | PASS. backup.yaml present (uploads ride the snapshot; postgres round-trips via --all-connected group/recover, DB-first); all images digest-pinned (glitchtip 6.2.3, valkey 8.1-alpine); migrations auto-apply staged-safe on start. |
| 6 Estate citizenship | PASS. Consumes estate postgres/smtp/oidc via protocol-named plugs; vault-autogen for secret_key/admin_password/postgres_password/keycloak_client_secret; no default creds; measured limits (web 1G with rationale, valkey 256M). Note: valkey is bundled as an in-stack broker (the paperless/medusa house pattern), not consumed via a redis plug, acceptable for a cache/broker. |
| 7 Honest health | PASS. Both services carry real healthchecks: glitchtip-web probes /_health/, valkey valkey-cli ping. “2/2 healthy” is a liveness claim. |
| 8 Maintained upstream | PASS. v6.2.3 (2026-07-31), commits today, not -classic (§2.4). |
Net: GlitchTip is a clean mechanical pass on all eight bars, and the only
owed item is the same from-empty live proof owed by every other candidate. It
is the correct occupant of the errors league.
4. The uptime/status league: split, don’t stack
Section titled “4. The uptime/status league: split, don’t stack”The task was whether GlitchTip should take BOTH errors and the uptime/status
league, or whether it should be errors = GlitchTip + uptime = uptime-kuma.
Head-to-head on the status capability (the league’s defining function):
| Capability | GlitchTip | uptime-kuma (apps/uptime-kuma) |
|---|---|---|
| Uptime checks | Ping/GET/POST/Heartbeat, project-scoped, alerting | HTTP/TCP/DNS/keyword/… many types |
| Public status page (anonymous, shareable) | NONE (issue #36 open since 2022, §2.3) | First-class: publishable status pages are a headline feature |
| Public-vs-private split | impossible (no anonymous surface) | native: status pages are separately publishable while the admin stays gated |
| Multi-user / SSO | native OIDC (MIT) | none, MIT single-admin, no SSO at any price; auth.disable + the SHC gate is the per-person path (landed) |
Recommendation: keep the split. The uptime/status league is fundamentally
about a public status surface, and GlitchTip structurally cannot provide
one, so awarding it both leagues would put a claim on a capability the product
does not have. uptime-kuma owns that capability outright. Operators are best
served by running BOTH: GlitchTip for private, OIDC-gated error tracking (with
its internal uptime alerting as a bonus), and uptime-kuma for the public status
page. This mirrors the DNS/function doctrine: errors and uptime are
different functions, and one product should not hold a league for a feature it
lacks.
If the owner instead defines the league as pure internal uptime alerting (no public page), GlitchTip could hold both, but that redefines the league around GlitchTip’s limitation rather than the capability operators ask for (“give me a status page I can share”), so this doc recommends against it.
Owed before uptime-kuma can wear the uptime label (not done here; labels
are owner-assigned): uptime-kuma has no upstream user model, so its flagship
rung is userless-but-gated, which requires authoring the flagship-tier.md
§2 grammar in apps/uptime-kuma/app.yaml: a top-level access: gated PLUS a
# flagship-userless: marker. Its app.yaml today argues the posture in prose
(the auth.disable block) but declares neither key. That is the whole gap for
the uptime-kuma-as-flagship path.
5. What this change ships (apps/glitchtip)
Section titled “5. What this change ships (apps/glitchtip)”Canonical layout, modeled on apps/paperless (the django/allauth sibling):
app.yaml: typed vars,access: open(see below),glitchtip-inithooks.compose.yaml:glitchtip-web(imageglitchtip/glitchtip:6.2.3,SERVER_ROLE=all_in_one: one image runs web + embedded django-vtasks worker- scheduler; not Celery) and an in-stack
valkeybroker. Digest-pinned.
- scheduler; not Celery) and an in-stack
vars.yaml: vault-autogen secrets;postgres/smtp/keycloakblocks.postgres.plug.yaml: estate DB (required unless a hostname is set); GlitchTip needs PostgreSQL 14+ (https://glitchtip.com/blog/2025-05-19-glitchtip-5-0-released/), the estate percona-16 store is in range.oidc.plug.yaml: the zero-touch Keycloak SocialApp seed (§2.2) + a matchingunintegrationcleanup, both verified read-back.smtp.plug.yaml: feeds the individualEMAIL_*env (notEMAIL_URL, whose URL form breaks on an@in the username).backup.yaml,glitchtip-init.job.yaml,tests/tests.bats(render-level + web-answers).
Access posture: access: open (ungated), and it must stay that way.
GlitchTip carries its own auth AND ingests error events over public DSN
endpoints (/api/<project>/store/, /api/<project>/envelope/) authenticated
by a per-project DSN key, not a user session. A forward-auth gate would 401
every SDK and silently break ingestion, the whole product. Errors stay private
because GlitchTip auth-gates its dashboard/API itself, exactly the native-OIDC
posture recorded for huly (flagship-tier.md §7).
Gates: validate-templates clean (373/373), scripts/arch-fitness.sh clean,
shfmt -d -i 4 -ci clean. No flagship: label applied.
6. Recommendation to the owner
Section titled “6. Recommendation to the owner”- Assign
errors→ glitchtip (addflagship: errorstoapps/glitchtip/app.yaml), superseding highlight’s ruled-out candidacy. - Assign
uptime→ uptime-kuma after authoring its userless-but-gated marker; do NOT awarduptimeto glitchtip. - Treat GlitchTip’s from-empty install+login proof as the one owed item, identical in status to every other flagship candidate.
7. Verified vs assumed
Section titled “7. Verified vs assumed”Verified this session (primary sources, URLs + 2026-08-06 access date
inline): MIT license; the license key gates no feature; the four-provider-only
SOCIALACCOUNT_PROVIDERS env limitation and the DB-SocialApp route; allauth’s
per-request OIDC callback route (no restart); uptime monitor types; the ABSENCE
of a public status page (issue #36 open); v6.2.3 (2026-07-31) + same-day commits +
single maintainer; the 3-service compose and django-vtasks (not Celery);
individual EMAIL_* support; PostgreSQL 14+ floor; uptime-kuma’s public status
pages + auth.disable + no-SSO (read from apps/uptime-kuma). Package gates
(validate-templates, arch-fitness, shfmt) run green in the worktree.
Assumed / owed: GlitchTip’s zero-touch OIDC + admin bootstrap are verified
at the RENDER + logic + upstream-behavior level, not proven with a live
from-empty install and browser login (no keycloak/estate in this session),
the same gap every flagship candidate carries. The healthcheck /_health/
path, the manage.py seed against the live schema, and the DSN-ingestion
reachability test are asserted by tests/tests.bats but were not executed live
here.