.gitbay/wiki/Admin.org
694 lines · 34995 bytes
1#+title: gitbay admin guide
2
3One static binary (=gitbayd=), one SQLite file, bare repositories on
4disk, and the system =git=. Schema migrations run automatically on
5startup and on every admin command.
6
7* Install
8
9Build from source (=go build ./cmd/gitbayd=), install via the vanity
10module path (=go install gitbay.org/gitbay/cmd/gitbayd@latest=), or use
11a release build: =deploy/release.sh <tag>= cross-compiles reproducible
12linux/amd64, linux/arm64, and darwin/arm64 binaries with a SHA256SUMS
13manifest (CGO off, trimpath, stripped — byte-identical per commit and
14toolchain).
15
16#+begin_src sh
17install -m 755 gitbayd /usr/local/bin/
18adduser --system --group --home /var/lib/gitbay --shell /usr/sbin/nologin gitbay
19install -d -o gitbay -g gitbay -m 750 /var/lib/gitbay
20gitbayd --config /etc/gitbay/config.toml check-config
21#+end_src
22
23=deploy/= in the source tree has a cloud-init file, a hardened systemd
24unit, and a nightly backup timer. Run as the unprivileged =gitbay= user;
25the unit's =AmbientCapabilities=CAP_NET_BIND_SERVICE= covers ports
2622/80/443 without root.
27
28** The SSH port decision
29
30- =ssh.mode = "embedded"= (default): gitbayd itself listens, normally on
31 22 — move the host's admin sshd to another port. Remotes read
32 =git@host:owner/repo= with no port gymnastics.
33- =ssh.mode = "system"=: the host sshd owns 22 and invokes gitbayd via
34 =AuthorizedKeysCommand=:
35 #+begin_example
36 AuthorizedKeysCommand /usr/local/bin/gitbayd --config /etc/gitbay/config.toml authorized-keys %t %k
37 AuthorizedKeysCommandUser gitbay
38 #+end_example
39 sshd requires that binary to be root-owned and not group/world
40 writable. Unknown keys fail authentication inside sshd, so system mode
41 requires =registration.mode = "closed"= (check-config enforces this).
42
43* Configuration reference
44
45=/etc/gitbay/config.toml=. =check-config= validates and names every
46contradiction; =--no-host-checks= skips port/path probes.
47=gitbayd admin config show= prints the configuration in effect as TOML,
48every default filled in and =smtp_pass= redacted. A file that fails
49validation still prints, followed by the contradiction.
50
51** [server]
52- =root= (default =/var/lib/gitbay=) — repositories, database, host
53 keys, ACME cache all live here.
54- =site_url= (required) — canonical =https://host=; drives ACME, clone
55 URLs, mail links.
56- =source_repo= (optional, =owner/name=) — the repository this instance
57 develops itself in. Startup warns when the running build's commit is
58 not on that repository's default branch, which is how a binary built
59 from an unmerged branch stops being invisible. Leave it unset unless
60 the instance hosts its own source.
61
62** [ssh]
63- =mode= — =embedded= | =system= (above).
64- =port= (22) — embedded listener port.
65- =host_keys= — list of private key paths; empty generates an ed25519
66 key at =<root>/ssh/host_ed25519=.
67
68** [http]
69- =addr= (=:443=), =tls= — =acme= | =files= | =off=.
70- =acme=: certificates via TLS-ALPN-01 on the HTTPS port, cached at
71 =<root>/acme=; =acme_email= for the CA account; =acme_http_addr=
72 (=:80=, ="off"= to disable) adds HTTP-01 and an https redirect —
73 failing to bind it is a warning, not fatal. Requires an =https://=
74 site_url with a public DNS name.
75- =files=: =cert_file= + =key_file=.
76- =off=: plain HTTP — development, or behind a TLS-terminating proxy.
77
78=trusted_proxies= lists the addresses or CIDRs of reverse proxies in
79front of the daemon. A request from one of them is attributed, for API
80rate limiting, to the last =X-Forwarded-For= hop that is not itself a
81trusted proxy; from anyone else the header is ignored. Empty, the
82default, is right when gitbayd terminates TLS itself.
83
84** [web]
85- =mode= — =view_only= (default) | =accounts=. In view_only the mutating
86 web routes are never registered; in accounts, browser sessions are
87 minted over SSH (=web login=), and users with write access can create
88 repos, comment, and make simple file edits (which commit unsigned,
89 honestly). =password_auth= is reserved and currently rejected.
90- =title= — the instance's display name in the rail and page titles;
91 empty falls back to the site host. Lower case is the convention for
92 gitbay itself.
93- =privacy_notice= — operator text shown on =/privacy= under the fixed
94 statement.
95
96** [registration]
97- =mode= — =closed= (default) | =invite= | =open=. invite/open require
98 [mail]. See the user guide for the flows.
99
100- =pending_expiry= (empty, never) — a duration such as ="168h"=; a
101 self-registered account still unverified after that long is removed,
102 hourly and at start, audited as =pending.expired=.
103
104** [mail]
105- =smtp_host= (host:port, 587 assumed), =from=, optional =smtp_user= /
106 =smtp_pass=. STARTTLS when offered. Required for invite/open
107 registration and self-service =email add=; in closed mode you may omit
108 it entirely and assert addresses by hand (below).
109
110** [api]
111- =enabled= (false) — the JSON API surface; see [[API]]. Off
112 means no credential-bearing HTTP endpoint exists at all.
113
114** [webhooks]
115- =allow_local= (false) — permit webhook targets on loopback/private
116 addresses. Leave off unless you know why you need it (SSRF).
117
118** [limits]
119- =clone_timeout= (3600s) — cap on =repo import= fetches.
120- =max_blob_bytes= (100MB) — cap on raw file serving over the web.
121- =max_asset_bytes= (512MB) — cap per uploaded release asset.
122- =max_snippet_bytes= (1MB) — cap per snippet file.
123- =max_snippets_per_user= (0, unlimited) — snippets an account may own.
124- =max_repos_per_user= (0, unlimited) — repositories an account may own
125 directly; =repo create=, =fork= and =import= refuse past it.
126 Organizations are not capped.
127- =max_bytes_per_user= (0, unlimited) — disk the account's own
128 repositories may take; a push may be no larger than what is left.
129- =max_pack_bytes=, =ssh_auth_rate= — reserved, not yet enforced.
130
131** [git_daemon]
132- =enabled= (false), =port= (9418) — the anonymous =git://= listener.
133 Serves only public repositories that additionally ran
134 =repo settings git-daemon <repo> on=.
135
136** [mirrors]
137- =pull_interval_minutes= (15) — how often pull mirrors fetch their
138 upstream. Push mirrors sync shortly after each local ref update.
139 Mirror URLs pass the same SSRF rules as webhook targets.
140
141** [go_import]
142Vanity Go module paths, one per line: ="host/module" = "owner/repo"=.
143Requests with =?go-get=1= at or under the module path answer with the
144go-import meta tag pointing at the repository's HTTPS clone URL, so
145=go install host/module/cmd/...@latest= resolves. The repository should
146be public (the module path itself confirms it exists).
147
148* Users, email, invites
149
150Every =gitbayd admin= subcommand except =backup=, =gc= and the one-shot
151backfills is a wrapper that dispatches the registry command of the same
152name as the host: an admin context with no account behind it, so its
153audit rows carry no actor and =source: host=. The same commands run in an
154instance admin's SSH session (=ssh git@<host> admin ...=) and write the
155same rows with the key fingerprint as source. One implementation, two
156credentials.
157
158#+begin_src sh
159gitbayd admin user create alice --key alice.pub --email a@example.org --verified [--admin]
160gitbayd admin email verify alice a@example.org # admin assertion, no SMTP needed
161gitbayd admin invite --email b@example.org # mails a code; prints it if no SMTP
162#+end_src
163
164"Verified" means SMTP-confirmed or host-admin-asserted; the database
165records which. Verified emails are what make commit signatures
166meaningful — an unverified address never produces a =verified= badge.
167
168* Audit and account control
169
170The audit log is the security feed (events are the product feed): every
171successful mutating command with its argv and source credential (SSH key
172fingerprint or API), registrations, admin actions, force-pushes, and
173auth failures/throttling. Secrets never appear — they travel on stdin,
174never in argv.
175
176#+begin_src sh
177gitbayd admin audit [--actor u|-] [--action prefix] [--since 24h|7d|date] [--limit n] [--json]
178ssh git@<host> audit ... # the same, from an admin session (SSH only)
179ssh git@<host> admin user list [--state active|pending|disabled|admin]
180ssh git@<host> admin user show <name> # keys, emails, orgs, tokens, sessions
181ssh git@<host> admin user limits <name> [--repos n|default] [--bytes n|default] # per-account caps
182ssh git@<host> admin user promote <name> # grant instance admin
183ssh git@<host> admin user demote <name> # remove it; the last admin is refused
184gitbayd admin user promote <name> # host-local: recovery when no admin key is reachable
185gitbayd admin user disable <name> # suspend: SSH, web, API all refused;
186gitbayd admin user enable <name> # sessions dropped, nothing deleted
187gitbayd admin user delete <name> --yes # only for accounts anchoring nothing:
188 # refused (with each blocker named) while
189 # the account owns repos, authored
190 # issues/MRs/comments/reviews, or is an
191 # org's only admin
192#+end_src
193
194=--actor= takes a username, or =-= for rows with no actor: host commands
195and auth failures. =--action= is a prefix, so =cmd repo= catches every
196repository command and =admin= every host or admin-session action.
197=--since= is a duration back from now (=30m=, =24h=, =7d=) or a date.
198
199=admin user list= pages by username (=--limit=, =--cursor=) and carries
200each account's state and =last_seen=, the newest use of any of its SSH
201keys or API tokens. =admin user show= adds the keys with their last use,
202each address with how it was verified, PGP keys, org roles, the owned
203repository count, API token names, and live browser sessions. Both are
204SSH-only and refused to non-admins, like =audit=.
205
206Promotion needs an active account: a pending or disabled one is refused.
207Demotion is refused when it would leave no admin, over SSH and on the
208host alike, so the host-local =promote= is the way back in when the only
209admin key is lost.
210
211Instance admin carries no right on anyone's repository: policy does not
212consult it, and a private repository still answers not-found to an
213admin. Moderation goes through explicit overrides that skip the access
214check and write their own audit row:
215
216#+begin_src sh
217ssh git@<host> admin repo list [--owner o] [--visibility public|private] # size, last push
218ssh git@<host> admin repo archive|unarchive <owner/name>
219ssh git@<host> admin repo visibility <owner/name> public|private
220ssh git@<host> admin repo delete <owner/name> --yes
221#+end_src
222
223Each lands in the audit log as =admin repo.<action>= naming the
224repository, on top of the =cmd= row every mutating command gets.
225
226=limits.ssh_auth_rate= (10) throttles per-IP authentication *failures*
227per minute — successful auths never count and clear the slate.
228=limits.max_pack_bytes= is enforced as =receive.maxInputSize= on every
229push.
230
231=limits.write_rate= (60) bounds *mutating commands per account per
232minute*. It is counted in the dispatcher, so SSH, the JSON API and the
233web spend one budget and a caller cannot refresh it by changing surface;
234=limits.api_rate= stays in front of it, bounding a network source rather
235than an account. A command is one token whatever it writes, so a bundle
236import costs one and only a loop of separate commands spends the budget.
237Read-only commands, the runner protocol (a build streams its log in many
238small writes) and the host CLI are exempt. Refusals exit 4 and say when
239to retry. A negative value turns the limit off; it matters most with
240=registration = "open"=, where every write also queues notification mail
241and webhook deliveries.
242
243* Queues
244
245Every background worker keeps a backlog and a failure state. An instance
246admin reads them all in one place:
247
248#+begin_src sh
249gitbay dashboard --json | jq .queues # webhooks, mail, mirrors, builds, deps
250#+end_src
251
252Per worker: pending, retrying (pending with a failed attempt) and
253dead-lettered counts with the oldest pending age, and the retrying or
254failed rows themselves, capped at twenty each. Builds list what is
255running and then what is pending, each since when, so a build no runner
256is scoped to claim is visible here rather than only in its repository;
257mirrors list the ones whose last sync failed;
258dependency checks list the ones whose last check errored. Non-admins get
259no =queues= key at all.
260
261In accounts mode the same read renders at =/admin=, linked from the rail
262for admins. Anyone else gets a 404 there.
263
264A dead-lettered mail is logged as =notification dead-lettered mail=<id>=,
265with the address redacted out of the relay's error. The id is the queue
266row: find it in the Mail table on =/admin=, or in =dashboard --json=,
267where the recipient and the unredacted error are. That is deliberate —
268see the Threat-Model page.
269
270* Maintenance
271
272#+begin_src sh
273gitbayd admin stats [--json] # counts, database size, per-repo disk
274ssh git@<host> admin stats [--json] # the same, from an admin session
275gitbayd admin gc [--repo owner/name] # git gc: repack and prune; per-repo sizes
276gitbayd admin gc --aggressive # thorough repack; slow, rarely needed
277gitbayd admin gc --lfs # also drop LFS objects no pointer names (older than a day)
278#+end_src
279
280A history rewrite leaves the commits it removed reachable through
281=refs/merge-requests/N/head= of the merge requests that landed them, so
282they stay fetchable by anyone who can read the repository. Nothing drops
283a head ref on its own — an open or source-gone MR is merged through it,
284and a merged or closed one keeps its diff readable through it — so the
285cleanup is a command an instance admin runs, naming the MRs:
286
287#+begin_src sh
288ssh git@<host> admin mr prune owner/name 1 2 3 --yes
289#+end_src
290
291It refuses an open or source-gone MR, deletes the named refs, runs
292=git gc --prune=now= on that one repository so the objects go at once
293rather than after git's two-week grace, leaves a system comment on each
294MR, and audits as =admin mr.prune=. The MR keeps its title, comments,
295reviews and head sha; =mr diff= and the MR page say the head is gone.
296Run it when nothing is pushing to that repository: without the grace, a
297push caught between leaving quarantine and writing its ref loses its
298objects. Objects also survive in offsite backups until those are pruned.
299
300=deploy/cloud-init.yaml= ships a =gitbay-gc.timer= that runs =admin gc=
301weekly (Sunday 07:00 UTC). Imported repositories keep whatever pack
302layout the source sent, so a first manual =admin gc= after a bulk
303import is worthwhile.
304
305* Backup and restore
306
307#+begin_src sh
308gitbayd admin backup --out /var/backups/gitbay/backup.tar.gz
309gitbayd admin backup --verify /var/backups/gitbay/backup.tar.gz # read it back
310#+end_src
311
312One archive: a consistent SQLite snapshot (taken *before* the
313repositories are read, so the database never references objects the
314archive missed), every repository, and the SSH host keys. Excluded:
315hook socket, regenerated hook scripts, WAL files. Safe to run against a
316live daemon.
317
318=--verify= reads an archive back: the snapshot must pass SQLite's
319integrity check, and every repository the snapshot names must be in the
320archive. A database-only archive is checked for integrity and says so.
321Exit is non-zero on damage or a missing repository.
322
323Restore: extract into an empty directory, point =server.root= at it,
324start gitbayd. Host keys are preserved, so clients keep their
325known_hosts entries; hooks regenerate at startup.
326
327** Schedule and recovery point
328
329Two timers, because the two halves of the data have different exposure.
330
331- =gitbay-backup.timer=, nightly. The full archive above, last 7 kept.
332- =gitbay-db-backup.timer=, hourly. =admin backup --db-only=, which
333 writes the SQLite snapshot alone, last 48 kept. A few MB against the
334 full archive's hundreds, which is what makes the frequency affordable.
335
336The split follows what a loss would actually cost. Repositories are git,
337so a mirror or any clone is a second copy; the database is the only copy
338of issues, merge requests, comments and review state. So the recovery
339point is about an hour for the data that exists nowhere else, and a day
340for the data that does.
341
342Continuous replication (litestream and similar) was considered and not
343adopted. It would take the database's recovery point to seconds, but the
344repositories would still be on the nightly archive, so a restore could
345produce a database referencing commits the repository backup does not
346have. Consistency between the two halves is worth more here than latency
347on one of them. Revisit if repository replication becomes continuous
348too.
349
350* Upgrades
351
352Replace the binary, restart the unit. Migrations apply automatically and
353are transactional; hook scripts under =<root>/hooks= are rewritten at
354startup to point at the current binary path.
355
356* CI runner
357
358=gitbay-runner= executes builds queued by pushes and merge requests. It
359polls over SSH with a key of scope =runner=, which reaches only the
360runner protocol and read-only git (a runner executes arbitrary
361repository code, so the key it holds must not do more). A runner key
362claims builds only for the repositories it is attached to, by =repo
363runner add= from a repository admin or an instance admin; an admin key
364claims any. Users attach their own runners: see the Users page. For an
365instance runner, run it as a dedicated unprivileged user on a non-admin
366account. =admin user create --key= registers a full-scope key, so the
367runner key is added afterwards through a bootstrap key that is then
368removed, and attached to each repository it should build:
369
370#+begin_src sh
371useradd --system --create-home --home-dir /var/lib/gitbay-runner ci-runner
372sudo -u ci-runner ssh-keygen -t ed25519 -N "" -f /var/lib/gitbay-runner/.ssh/id_ed25519
373ssh-keygen -t ed25519 -N "" -f /tmp/ci-bootstrap
374gitbayd --config /etc/gitbay/config.toml admin user create ci --key /tmp/ci-bootstrap.pub
375ssh -i /tmp/ci-bootstrap git@127.0.0.1 keys add --scope runner < /var/lib/gitbay-runner/.ssh/id_ed25519.pub
376ssh -i /tmp/ci-bootstrap git@127.0.0.1 keys remove "$(ssh-keygen -lf /tmp/ci-bootstrap.pub | awk '{print $2}')"
377rm /tmp/ci-bootstrap /tmp/ci-bootstrap.pub
378gitbay-runner -remote git@127.0.0.1 -workdir /var/lib/gitbay-runner/work
379#+end_src
380
381#+begin_src sh
382gitbay repo runner add krz/site < /var/lib/gitbay-runner/.ssh/id_ed25519.pub
383#+end_src
384
385=-jobs N= runs N builds at once. Claiming is one transaction that
386selects and updates, and each build works in its own =build-<id>=
387directory, so workers do not collide; idle polls are staggered across
388the interval so N of them do not wake together. The drop-in's weights
389below are per service, not per build, so raising =-jobs= divides them
390rather than multiplying the host's load.
391
392=admin runners= shows which account each runner polls as, and what each
393may claim. A runner key claims builds only for the repositories it is
394attached to: with none attached it claims nothing, and =-repos= may only
395narrow within them. An admin's full-scope key claims any repository —
396that is what =-repos= was for — and still works for the protocol during
397a rotation. A merge request head from a fork is built in the target
398repository as untrusted: the claim carries no secrets, and only a runner
399started with =-untrusted= takes it. Same-repository heads were built by
400their branch push and are not built again.
401
402=make deploy-runner= also installs
403=deploy/gitbay-runner.override.conf= as a systemd drop-in: =Nice=10=,
404=CPUWeight=30=, =IOWeight=30=, so a build never starves the host's sshd,
405the daemon or the backup timers, and =NoNewPrivileges=,
406=ProtectSystem=full=, =ProtectKernelTunables=, =ProtectControlGroups=
407and =RestrictSUIDSGID=, so a step cannot reach outside its workspace
408and the runner's home. The e2e suite alone starts sixty
409daemon instances; without the drop-in a deploy's copy over the admin
410sshd stalled. Both deploy targets copy with =rsync --partial=, which
411resumes a stalled transfer.
412
413Under =-isolation podman=, the default and what bay1 runs, each build
414is confined to a container (see Container isolation below). Under
415=-isolation none= steps run directly on the host as the runner's user,
416so treat that machine as executing whatever your users push, and
417install the toolchains your builds need on it.
418
419A runner claims the oldest pending build among the repositories its key
420is attached to — for an admin key, the oldest in the instance. =-repos=
421narrows within that set, which is what makes a runner outside the server
422practical: one on a machine that should build a single project, or that
423holds credentials for one deployment, stays on it.
424
425Oldest-first is across everything the key may claim, so a repository
426with a deep queue holds every other repository the same runner serves;
427bay1 measured a 15-minute average wait on a day of merge request
428stacks from one repository. A runner attached to one repository cannot
429be starved. That is the rule, decided in krz/gitbay#207: a runner
430serving several repositories takes them oldest-first, and an operator
431who wants one repository never to wait on another runs a second
432runner attached to it alone. Nothing caps what an account queues,
433and nothing needs to (krz/gitbay#206): a schedule tick queues nothing
434while the job's last build is pending or running, so a repository
435with no runner holds one row per scheduled job rather than one per
436tick, and a build runs only on a runner its owner attaches, so a busy
437schedule spends the owner's compute. Pushes are bounded by what an
438account can push.
439
440#+begin_src sh
441gitbay-runner -remote git@gitbay.org -repos krz/site,krz/docs \
442 -workdir /var/lib/gitbay-runner/work
443#+end_src
444
445Add =-untrusted= only with =-isolation podman=.
446
447gitbay.org's runner is attached to the forge's own repositories and
448the isolation canary, nothing else, because it shares the host with
449the forge; its unit names no =-repos=, the attachments are the
450boundary. Any other repository builds on a runner its owner attaches.
451
452=-repos= narrows an admin runner; for a runner key the attachments are
453the boundary, held by the server, and =-repos= may only name
454repositories among them. =-untrusted= makes a runner claim merge
455request heads from forks; the bay1 unit sets it because it isolates in
456podman. A runner without it builds trusted commits only.
457
458=gitbay dashboard= and =ssh git@<host> admin runners= list every key
459that has polled as a runner: the account, the key's fingerprint, when it
460last polled, the repositories it may claim — its attachments for a runner
461key, the =-repos= it asked for or =any= for an admin key — and the build
462it holds; =admin runners remove <fingerprint>= (=forget= until the next release) drops the row for a key
463that polled by mistake, the key itself untouched. =admin runners= also
464heads the list with the queue: builds
465pending now, and over the last day how many were claimed, how long they
466waited to be claimed (average and worst), and
467how many the reaper ended instead of a runner reporting them. A build a runner claimed and never
468reported is failed by the scheduler's minute tick, whether or not any
469runner is still alive: within about two minutes of its log stream ending
470with no outcome reported — the runner reports right after closing the
471stream, retrying for half a minute if gitbayd is unreachable — or, if no
472stream was ever seen, at the deadline.
473
474Instance admin on the runner account only authorizes the claim/report
475protocol; it grants no repo access. A build that pushes back — a pages
476deploy, an archive publish, an automated MR branch — needs an explicit
477grant on that repo: =repo access grant <owner/name> ci write=. Private
478repos likewise need at least read for the clone.
479
480** Container isolation
481
482Builds run in a rootless podman container, one per job, with the
483workspace bind mounted and nothing else: the clone happens outside with
484the runner's key, so a step cannot read it. =-isolation none= keeps the
485old behaviour — steps on the host as the runner's user — for an instance
486where every repository is trusted. There is no automatic fallback: a
487runner started with =-isolation podman= that cannot find a working
488podman exits rather than running a build unsandboxed.
489
490gitbay's own jobs name =localhost/gitbay-ci:2=, built from
491=deploy/Containerfile.ci= on the runner host. A job's image must carry
492what its steps need: the suite drives real git, git-lfs, gpg and sshd and
493asserts they exist before running, so the stock runner default would fail
494it immediately. Build or rebuild it with:
495
496#+begin_src sh
497ssh -p 2222 root@<host> 'cat > /tmp/Containerfile.ci' < deploy/Containerfile.ci
498ssh -p 2222 root@<host> 'su - ci-runner -s /bin/sh -c \
499 "podman build -t localhost/gitbay-ci:2 -f /tmp/Containerfile.ci /tmp"'
500#+end_src
501
502The tag is deliberate rather than =:latest=: changing the file means
503bumping the tag in =.gitbay/ci.yml=, so a running branch's image does not
504change under it.
505
506=-image= names the image a job runs in when it declares none, and is
507required under =-isolation podman=: there is no built-in default,
508because an image this host does not have would fail every build. A job
509overrides it with =image:= in =.gitbay/ci.yml=, validated as a reference
510so a config file cannot turn it into podman arguments.
511
512=-cpus= and =-memory= cap one build (podman's units, e.g. =-cpus 2
513-memory 4g=); unset means uncapped. The runner applies them itself: it
514creates a cgroup per build under its own delegated service cgroup,
515writes the limits there, and starts every podman process for the build
516inside it, with podman's cgroup handling off. Podman's own =--memory=
517and =--cpus= never applied under rootless cgroupfs, which is what a
518system service gets (krz/gitbay#188). The unit therefore needs
519=Delegate=yes=, which the drop-in sets; without it the runner refuses
520to start when a limit is set, and logs that builds run unconfined when
521none is. bay1 runs =-cpus 3 -memory 6g= per build inside =MemoryMax=6G=
522and =CPUQuota=300%= on the unit, on a 7.7GB four-core host with no
523swap: the memory cap is what keeps the forge alive when a build
524allocates without bound, and it sits above the e2e suite's 5GB peak
525rather than at a fair share. =OOMPolicy=continue= keeps systemd from
526stopping the runner when a build is OOM-killed.
527
528Each repository gets its own build home under the runner's workdir,
529mounted into its containers as =HOME=. Caches persist between builds of
530one repository and are never read by another's.
531
532*Images are provisioned, never pulled by a build.* The runner passes
533=--pull=never=. Two reasons, and the second is the better one: the
534service runs with =RestrictSUIDSGID=yes= so podman cannot unpack a layer
535holding a setuid file, which is nearly every distribution image; and on
536an instance where anyone can push a =ci.yml=, =image:= would otherwise
537mean "fetch and run anything from the internet". An operator pulls or
538builds what is allowed and a build picks among those. A job naming an
539image the host does not have fails with a message saying so.
540
541#+begin_src sh
542su - ci-runner -s /bin/sh -c "podman pull docker.io/library/alpine:3.20"
543su - ci-runner -s /bin/sh -c "podman images"
544#+end_src
545
546Prepare a host before pointing an isolating runner at it:
547
548#+begin_src sh
549ssh -p 2222 root@<host> 'sh -s' < deploy/runner-podman-setup.sh
550make deploy-runner
551#+end_src
552
553The script installs podman, delegates a subuid/subgid range to
554=ci-runner=, checks that user namespaces are enabled rather than
555assuming, enables lingering, and verifies rootless podman actually runs
556as that user. It is idempotent.
557
558The drop-in sets =NoNewPrivileges=no=, without which rootless podman
559cannot call =newuidmap= and the runner refuses to start. That is a
560considered trade, explained in the file and in the Threat-Model; if you
561run with =-isolation none=, set it back to =yes=.
562
563*Restarting the runner is safe.* On SIGTERM it stops claiming, finishes
564the build in flight, reports it, and exits; the drop-in's
565=TimeoutStopSec=50min= covers the longest build, and its =KillMode=mixed=
566is what makes the signal reach the runner alone — under systemd's default
567the build's container and the log session are signalled with it, and the
568runner drains a build that is already dead. So =make deploy-runner=
569waits for a running build rather than orphaning it, and a build's result
570is retried for half a minute if gitbayd is restarting at that moment. A
571second SIGTERM ends the runner at once, abandoning the build to the
572reaper. The suite checks all three: =TestRunnerDrainsOnSIGTERM= signals
573the process, =TestRunnerDropInLetsTheDrainHappen= reads the drop-in's
574=KillMode= and =TimeoutStopSec=, and =TestRunnerDrainUnderSystemd= runs
575the runner as a transient user unit under =systemd-run= and stops it
576under both kill modes. That last one needs a systemd user manager, so it
577skips in the container CI runs in; run it on a Linux host with
578=go test ./e2e -run TestRunnerDrainUnderSystemd -v=.
579
580*Validate podman mode on a scratch repository before pointing the runner
581at real ones.* Every deploy that switched the whole instance to
582containers and failed took CI down with it. Instead: create a throwaway
583repository the runner account can read (public, or granted read — a
584private one is "not found" to the runner and the build stays pending),
585give it one job that names the CI image, and deploy the runner with
586=-repos= naming only that repository. The production unit, with its real
587hardening, then claims nothing else; other repositories' builds queue
588until =-repos= is switched back, which is a pause, not an outage.
589
590#+begin_src sh
591gitbay repo create cmc/ci-smoke # then push a .gitbay/ci.yml naming the image
592sed -i 's#-repos krz/gitbay #-repos cmc/ci-smoke #' /etc/systemd/system/gitbay-runner.service.d/override.conf
593systemctl daemon-reload && systemctl restart gitbay-runner
594gitbay build log cmc/ci-smoke 1 # green: switch -repos back, redeploy
595#+end_src
596
597*Do not deploy an isolating runner to a host that has not been
598prepared.* The runner is specified to refuse to start without a working
599podman rather than fall back to running builds unsandboxed — a fallback
600that silently drops isolation is worse than a stopped runner, because
601nothing surfaces it. On an unprepared host that refusal stops every
602build on the instance.
603
604The service drop-in carries =Delegate=yes= for rootless cgroup
605management and =ReadWritePaths= for podman's store under
606=/var/lib/gitbay-runner=, which =ProtectSystem=full= would otherwise
607make read-only. Those paths are prefixed =-= so they are ignored when
608absent: the drop-in installs on unprepared hosts too, and a unit that
609refused to start would stop every build.
610The nightly canary on =cmc/ci-smoke= only runs if the runner's =-repos=
611names that repository too; a scoped runner claims nothing else.
612=gitbay-runner-prune.timer= prunes unused images weekly, as the runner's
613user: rootless storage belongs to that user, and root's prune would not
614see it. An unpruned image store on a 40GB host is a slow outage.
615
616* LFS storage
617
618Objects live content-addressed under =[lfs] root= (default
619=<server.root>/lfs=); =[lfs] max_object_bytes= caps a single object
620(512MB default). Storage sits behind a small interface — an
621S3-compatible backend is a drop-in with the server proxying, and
622presigned URLs a later optimization. LFS objects do not travel with
623push mirrors (mirrors move git refs only), and gc does not yet collect
624orphaned objects.
625
626* Pages
627
628=[pages] domain = "example.site"= serves public repos' =pages= branches
629on =<owner>.<domain>=. DNS needs a wildcard record =*.<domain>= to the
630server; ACME issues per-subdomain certificates on demand (only for
631owners that exist). The domain must not be the site host or a parent of
632it — pages content runs its own scripts and must stay off the forge's
633origin.
634
635Users with repo admin claim custom domains with =repo domain add=.
636Claims activate only after a DNS TXT challenge proves control of the
637domain (=repo domain verify=, audit-logged); pending claims serve
638nothing, get no certificates, and expire after 7 days. ACME issues
639certificates only for verified hosts, so stray DNS pointed at the
640server gets nothing.
641
642* Security
643
644The [[Threat-Model]] file is the reference for what the forge
645trusts and refuses to do. Operational checklist:
646
647- *Software checks.* =deploy/audit.sh= runs =go vet=, =govulncheck=
648 (the module list is deliberately short — review it on each release),
649 and a short fuzz pass over every attacker-facing parser (pkt-line,
650 commit, SSHSIG armor, OpenPGP key, SSH tokenizer). Run it before
651 tagging a release. CI's own =vuln= job runs =govulncheck= nightly
652 against main rather than per push, because =@latest= scans today's
653 advisory database and an advisory lands without anyone pushing;
654 =build trigger krz/gitbay vuln= runs it on demand.
655- *Web responses* carry a scripts-forbidden CSP, =X-Frame-Options:
656 DENY=, =nosniff=, =no-referrer=, and HSTS when TLS is on — no
657 configuration needed.
658- *Host sandboxing.* The systemd unit in =deploy/cloud-init.yaml= runs
659 gitbayd unprivileged with =ProtectSystem=strict=, =PrivateDevices=,
660 =LockPersonality=, =MemoryDenyWriteExecute=,
661 =SystemCallFilter=@system-service=, and =RestrictAddressFamilies= to
662 INET/INET6/UNIX. It keeps =CAP_NET_BIND_SERVICE= only, to bind 22/80/443.
663- *OS patches* apply via =unattended-upgrades= (security origins,
664 auto-reboot 04:30 if required).
665- *Admin sshd (2222)* is throttled by =MaxStartups=/=MaxAuthTries= and
666 watched by =fail2ban=; gitbayd's own port 22 is throttled by
667 =limits.ssh_auth_rate= (auth failures per IP per minute), and every
668 account's writes by =limits.write_rate=.
669- *Monitoring.* =gitbay-monitor.timer= writes a reading hourly to
670journald and, when =/etc/gitbay/monitor.url= exists, posts it to that
671webhook: disk, service, the daemon's own =/healthz= answer, certificate
672expiry, and the age of the newest full backup and database snapshot.
673It exits non-zero on an alert so the unit shows in =systemctl
674--failed=: a stopped service, =/healthz= not answering =ok=, disk ≥ 85%,
675a certificate under 21 days, a full backup over 25 hours old, or a
676database snapshot over 2 hours old.
677
678=GET /healthz= is unauthenticated and cache-free: whether the database
679answers and which commit serves, 503 when it does not.
680- *Database.* =gitbay.db= and its WAL live under =/var/lib/gitbay= (mode
681 0750, owned by =gitbay=). The nightly archive plus provider snapshots
682 are the recovery path; for tighter RPO, add continuous replication
683 (litestream) against the same file — it coexists with the WAL.
684
685* Odds and ends
686
687- deleting a fork marks MRs sourced from it =source_gone=; their diffs
688 remain viewable and mergeable because the target repo owns the
689 objects.
690- =refs/merge-requests/*= is server-owned and unpushable by clients;
691 only =admin mr prune= removes one (see Maintenance).
692- audit-relevant activity (issue/MR lifecycle, imports, pushes) lands in
693 the =events= table, which also feeds webhooks.
694- the daemon idles under 10MB RSS; the smallest VPS tier is adequate.