cc-lib changed JobData, ScopedJobStats and Job.Statistics from flat maps
to structs (a .Metrics map plus array-valued Groups) to represent
filesystem (and future interconnect) metric groups. Migrate all
construction, indexing and iteration to the new API across the archive
backends, metricstore query path, metric dispatcher, archiver, importer,
tagger, taskmanager, repository and API layers.
Semantics: DecodeJobStats now projects JobData.Groups into
ScopedJobStats.Groups, and the archiver derives per-filesystem node-scope
statistics into Job.Statistics.Groups. deepCopy, resampling and the
metric/scope filter are group-aware. The metricstore internal storage
(buffers, selector tree, checkpoint/parquet) is unchanged; all conversion
stays at the LoadData/archive-codec seam.
Note: requires the corresponding cc-lib release; bump the cc-lib
dependency version once tagged (a local go.mod replace was used during
development and is intentionally not committed).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Browser sessions are now server-side, stored in the SQLite database via
scs/sqlite3store (new `sessions` table, DB migration to version 12) instead
of gorilla/sessions client-side cookie storage. Only an opaque random token
is kept in the cookie; session data lives server-side and survives restarts.
Session middleware is wired as a hybrid to avoid buffering large responses:
scs.LoadAndSave on the login/logout write paths, and a non-buffering
read-only LoadSession middleware on the secured/config/frontend read paths
so the large GraphQL /query responses stream unbuffered. JWT-only APIs
(/api, /userapi, /api/metricstore) and static files are left unwrapped.
The session cookie Secure flag is now derived from the server config (set
when cc-backend terminates TLS itself); previously it was effectively never
set. The SESSION_KEY env var is removed as server-side tokens need no
signing secret. The dormant Bearer-JWT branch in the frontend urql client
is removed; the web UI authenticates GraphQL via the session cookie.
Closes#558
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Entire-Checkpoint: b51075f43cc7
Addresses the remaining medium findings from the second-pass audit:
- DoS hardening: bound GraphQL query cost with FixedComplexityLimit, and
reject non-positive items-per-page / page values so uint64 conversion
cannot underflow into an unbounded LIMIT/OFFSET. The -1 "load all"
sentinel stays valid for dashboards; REST now returns 400 for bad input.
- Security headers: add X-Content-Type-Options, X-Frame-Options,
Referrer-Policy and a conservative CSP (frame-ancestors/object-src/
base-uri) that hardens against clickjacking and base-tag injection
without restricting the self-hosted SPA's inline scripts.
- Stored XSS: render job.metaData.message as escaped text instead of
{@html ...} in Job.root and JobFootprint, preserving line breaks via
white-space: pre-wrap.
- SQL injection hardening: parameterize the tag-scope IN list and the
manager project subquery in CountTags instead of interpolating
user.Username / user.Projects (externally sourced via OIDC/LDAP).
- CSRF defense-in-depth: reject cross-site state-changing requests via
Sec-Fetch-Site, failing open for non-browser clients, on top of the
existing SameSite=Lax session cookie.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Entire-Checkpoint: de7d47a85c7c
A second-pass audit surfaced three severe issues missed by the previous
review, each a sibling code path of a bug class that was only partially
fixed before:
- auth: JWT session login (jwtSession.go) registered its authenticator
even when CROSS_LOGIN_JWT_HS512_KEY was unset, leaving an empty HMAC
key. golang-jwt verifies any HS256/HS512 signature against an empty
key, allowing unauthenticated admin token forgery. Init() now refuses
to register without a key, with a defense-in-depth empty-key guard in
the keyfunc.
- repository: metric names from GraphQL ([String!]) were interpolated
raw into json_extract(footprint, "$.<name>") SQL. SQLite parses
double-quoted strings as literals, enabling SQL injection by any
authenticated user. Validate metric names against ^[a-zA-Z0-9_]+$ in
jobsMetricStatisticsHistogram and buildFloatJSONCondition.
- metricstore: cluster/host line-protocol tags flowed unvalidated into
path.Join(RootDir, cluster, host) for checkpoint/WAL files, allowing
arbitrary file write outside the checkpoint root via NATS
(unauthenticated) or POST /api/write. Reject path-traversal sequences
in DecodeLine before the tags become path components.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Entire-Checkpoint: b57246993ec1
- Add migration 14: partial covering indexes WHERE job_state='running'
for user/project/subcluster groupings (tiny B-tree vs full table)
- Inline literal state value in BuildWhereClause so SQLite matches
partial indexes instead of parameterized placeholders
- Add per-request statsGroupCache (sync.Once per filter+groupBy key)
so identical grouped stats queries execute only once per GQL operation
- Parallelize 4 histogram queries in AddHistograms using errgroup
- Consolidate frontend from 6 GQL aliases to 2, sort+slice top-10
client-side via $derived
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Entire-Checkpoint: 5b26a6e5ff10
Adds composite covering indexes on (cluster, job_state, <group_col>, ...)
for user, project, and subcluster groupings to enable index-only scans
for status views. Drops subsumed 3-column indexes.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Entire-Checkpoint: 3d8def28e96e
moebiusbandandClaude Opus 4.8
Christoph Kluge