Parity sweep 3#2382
Merged
Merged
Conversation
Upgrade JS/TS UI dependencies to latest (gate: install + lint + build green): - AWS SDK clients + credential-providers 3.1070.0 -> 3.1079.0 - @sveltejs/kit 2.65.2 -> 2.69.1, svelte 5.56.3 -> 5.56.4, svelte-check 4.6.0 -> 4.7.1, vite 8.0.16 -> 8.1.3 - @tailwindcss/vite + tailwindcss 4.3.1 -> 4.3.2 - oxlint 1.70.0 -> 1.72.0, oxfmt 0.55.0 -> 0.57.0 - @testing-library/svelte 5.3.1 -> 5.4.2 Pin-backs (kept on latest v1, rejected v2 majors): - @connectrpc/connect + connect-web 1.6.1 -> 1.7.0 (not v2) - @bufbuild/protobuf 1.10.0 -> 1.10.1 (not v2) connect/protobuf-es v2 would require regenerating the committed dashboard_pb.ts / dashboard_connect.ts from proto sources (buf toolchain) - out of scope for a dependency bump. Source fix: - .oxlintrc.json: disable new pedantic rule unicorn/prefer-number-coercion introduced in oxlint 1.72.0 (flags pre-existing parseInt/parseFloat; its Math.trunc(Number(x)) auto-fix is not semantics-preserving). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Personal local settings must not sync to origin (they carry per-user permission overrides). Remove from tracking and ignore going forward. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Add a CI job running `go fix -diff ./...` (Go 1.26 built-in modernizers, fails on non-empty diff) so the tree stays modernized. Repo was already near-clean; only ec2 AttachVerifiedAccessTrustProvider had a stale suppressed loop, now slices.Contains. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…validation, response overrides Deep parity audit of the s3 service against aws-sdk-go-v2. Twelve genuine correctness defects fixed (no cosmetic churn): - SSE data loss on restart: StoredObjectVersion.EncryptionDEK/Nonce and StoredMultipartUpload.SSE were json:"-" while the ciphertext (Data) was persisted, so every SSE-S3/SSE-KMS object became undecryptable after a snapshot/restore, and in-flight multipart uploads silently completed unencrypted. Now persisted (SSE-C customer key stays request-scoped). - GetObject/HeadObject: implement the response-* header override query params (response-content-type/-disposition/-encoding/-language, response-expires, response-cache-control) — previously ignored; heavily used via presigned URLs. - PutBucketAcl: reject object-only canned ACLs (bucket-owner-read / bucket-owner-full-control) with 400 InvalidArgument; read and honour an AccessControlPolicy XML body (was silently ignored); pass it through on GET. - PutBucketReplication: require versioning=Enabled (InvalidRequest 400), as AWS. - Presigned URLs: reject X-Amz-Expires > 604800 s (7 days) with 400 AuthorizationQueryParametersError. - x-amz-storage-class: omit the header for STANDARD objects (AWS omits it). - GetObjectAttributes: return Last-Modified header; drop omitempty on ObjectSize so a 0-byte object still emits <ObjectSize>0</ObjectSize>. - PostObject response: XML-escape bucket/key via encoding/xml instead of raw string concatenation (keys may contain & < >). - ListMultipartUploads: echo UploadIdMarker in the response. Tests updated to match real AWS behaviour (STANDARD storage-class header omission; replication requires versioning) and regression tests added for the SSE snapshot/restore round-trip, canned-ACL rejection, presign expiry cap, and response-header overrides. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…ifecycle accuracy
Deep AWS-parity audit of the ec2 service (gopherstack-r0h), focused on Tags
and the instance attribute/lifecycle family:
- CreateTags/DeleteTags/DescribeTags previously only recognised ~9 resource
types (instance, sg, vpc, subnet, volume, igw, route-table, natgateway,
elastic-ip) via resourceExistsLocked/resourceTypeByID, so tagging AMIs,
snapshots, network ACLs, transit gateways, VPN/customer gateways, VPC
endpoints, launch templates, IPAM objects, and ~80 other resource types
this backend models incorrectly failed with InvalidParameterValue even
though the resource existed. Replaced with a comprehensive, AWS
ResourceType-accurate prefix table and per-family existence checks
(backend_resource_types.go), covering every independently-taggable
resource type the backend implements.
- ModifyInstanceAttribute silently discarded disableApiTermination,
disableApiStop, ebsOptimized, instanceInitiatedShutdownBehavior, and
sourceDestCheck ("accepted but not modelled beyond acknowledgment") — a
disguised stub. DescribeInstanceAttribute then hardcoded all of them back
to fixed values (sourceDestCheck wrongly defaulted false; AWS defaults
true). Now persisted for real: booleans land on the Instance struct,
sourceDestCheck is proxied to the primary ENI (its true AWS owner), and
RunInstances' own DisableApiTermination/InstanceInitiatedShutdownBehavior/
EbsOptimized launch-time parameters are wired the same way.
- TerminateInstances/StopInstances never enforced disableApiTermination/
disableApiStop, so a protected instance could always be torn down —
added the OperationNotPermitted error AWS returns in that case.
- Instance carried no StateReason/StateTransitionReason, so DescribeInstances
never surfaced why an instance stopped/terminated. Added <stateReason>
and legacy <reason> wire fields, populated on user-initiated stop/
terminate and cleared on start, matching AWS's Instance shape.
Found but not fixed (follow-up): DeleteVpc/DeleteSubnet (backend.go
DeleteVpc, DeleteSubnet) force-cascade-delete all dependents instead of
returning DependencyViolation like real AWS; changing this has a large
blast radius on existing cascade-delete test assumptions and is left for a
dedicated pass.
Gate: go build ./..., go vet ./services/ec2/..., go fix -diff ./services/ec2/...,
go test ./services/ec2/..., golangci-lint run ./services/ec2/... all green.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…nsact-update key-mutation gaps
Deep parity re-audit of a previously-swept, mature dynamodb service (op-by-op
against aws-sdk-go-v2 dynamodb types). Found and fixed three genuine defect
families rather than padding an already-solid implementation:
1. Query/Scan Select parameter: the emulator never enforced AWS's documented
restriction that Select can only combine with ProjectionExpression/
AttributesToGet when Select=SPECIFIC_ATTRIBUTES, never rejected
SPECIFIC_ATTRIBUTES without a projection, and never rejected
ALL_PROJECTED_ATTRIBUTES on a bare table scan/query. It also always
returned full Items for Select=COUNT, when AWS returns Count/ScannedCount
only ("Returns the number of matching items, rather than the matching
items themselves").
2. BatchWriteItem had no duplicate-key validation across a table's
WriteRequest list (BatchGetItem already had this for its Keys list). Real
AWS rejects a batch that targets the same primary key twice — via two
Puts, two Deletes, or a Put+Delete pair — with ValidationException
"Provided list of item keys contains duplicates". An existing test
(TestBatchWriteItem_ValidRequests_NotAffectedByValidation) asserted the
opposite for a Put(k1)+Delete(k1) batch; fixed the test to use disjoint
keys and added a dedicated regression test for the rejection.
3. TransactWriteItems' Update action never validated that its
UpdateExpression avoids key attributes, unlike plain UpdateItem. Since
updateIndexes() only ever adds/overwrites the new key's index slot and
never removes a stale one, an unvalidated key-mutating transactional
update would silently corrupt pkIndex/pkskIndex (leaving a dangling entry
under the old key pointing at the item's new state) — a real state
corruption bug, not just a missing validation.
Gates: go build ./..., go vet ./services/dynamodb/..., go fix -diff (empty),
go test ./services/dynamodb/... (all pass), golangci-lint run
./services/dynamodb/... (0 issues).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Keeps the spa dir tracked when built output is absent/gitignored. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…root SNS backend.go already implements deliverToLambdaSubscriptions and deliverToFirehoseSubscriptions (called from Publish), gated on b.lambdaBackend/b.firehoseBackend being non-nil, but SetLambdaBackend/ SetFirehoseBackend were only ever called from tests -- never from cli.go. In the running binary, subscribing a Lambda function or Firehose delivery stream to an SNS topic and publishing silently no-opped. Add wireSNSToLambdaFirehose, called alongside wireSNSToSQS in initializeServices, wiring: - SNS.SetLambdaBackend(lambdaBk) directly (lambda's InvokeFunction already satisfies sns.LambdaInvoker since InvocationType is a string alias). - SNS.SetFirehoseBackend via a new snsFirehosePutterAdapter, since firehose.PutRecordBatch takes a context but sns.FirehosePutter does not. - SNS.SetSQSSender via the existing sqsSenderAdapter, so failed Lambda/ Firehose deliveries with a RedrivePolicy reach the real subscription DLQ. This is separate from wireSNSToSQS, which wires the SNS->SQS subscription delivery path via a publish emitter; sqsSender only serves DLQ redelivery on failed Lambda/Firehose invocations, so there is no double-wiring. Add TestWireSNSToLambdaFirehose_EndToEndDelivery, which builds real SNS, Lambda, Firehose, and SQS in-memory backends, calls the same wireSNSToLambdaFirehose used by cli.go, and proves genuine delivery: a Firehose subscription's published message is flushed to S3, and a Lambda subscription's invocation failure (no Docker runtime in test) is redirected to the real SQS dead-letter queue via the wired SQS sender. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…permission persistence
RemovePermission read StatementId from query string; real SDK sends it as
URI path segment (/policy/{StatementId}) — route never matched, a disguised
stub no client could call. Fix ESM function-ARN parsing that dropped the
function name (kept only qualifier). Snapshot permissions map + rebuild
versionIndex/esmByFunctionARN on Restore (were lost across persistence).
Add Qualifier scoping, EventSourceToken/PrincipalOrgID fields.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…rsistence leaks ListInstanceProfilesForRole ignored RoleName + returned hardcoded empty; GetAccountAuthorizationDetails fabricated a fake v1 policy version. Fix HTTP status codes (NoSuchEntity 404, EntityAlreadyExists/DeleteConflict/Limit 409, were all 400). Percent-encode policy documents at wire boundary. Snapshot handler tags + comprehensive backend state (SSH keys, MFA links, access-advisor) that were dropped on restore. Apply tags-at-creation. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
YAML-frontmatter parity manifests record audit state (last_audit_commit, sdk_version, per-op/family wire/errors/state/persist status, gaps, leaks) so the next audit diffs the delta instead of rescanning. Template at services/_PARITY_TEMPLATE.md. Backfilled s3/ec2/dynamodb/lambda/iam from their sweep reports. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…fan-out, delivery leaks PublishBatch dropped per-entry MessageAttributes (wrong form-field prefix vs SDK serializer) — broke FilterPolicy matching. Lambda/Firehose/SQS now share one real signed envelope (was fabricated Signature); Firehose honors RawMessageDelivery. ReplayPolicy fans out to all protocols (was HTTP/SQS only). Persist + clean up topicMessageArchive; cap delivery observability buffers; lock the signer cert URL; fix copy-paste KMSOptInRequired error message. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…cy enforcement, persistence Query-protocol responses emitted two XML prologs (manual header + XMLBlob) — malformed XML masked by lenient parsing. FIFO messages requeued on visibility change/expiry appended to tail, letting newer same-group messages jump ahead — now reinserted by SequenceNumber. Enforce RedriveAllowPolicy (was validated, never checked). Persist fifoSeqCounter/hasActivity/lastPurgedAt. Export NoVisibilityTimeout sentinel. Centralize VisibilityTimeout range check. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…ter extraction, RejectedLogEventsInfo shape PutLogEvents no longer validates sequenceToken / returns InvalidSequenceToken / DataAlreadyAccepted — AWS deprecated tokens and never errors on them. Real per-event field extraction for $-referenced MetricValue (was fabricating 1.0); TestMetricFilter now computes ExtractedValues (was empty stub). Fix RejectedLogEventsInfo: TooOldLogEventEndIndex name + exclusive-end off-by-one. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…missing comparison operator PutMetricDataOutput fabricated UnprocessedMetricData + partial-success — real op has no such model; now validate-then-commit atomically. Parse MetricDatum Values/Counts weighted arrays (were silently dropped) w/ percentile expansion. Reject NaN/Inf/out-of-range values. Add missing LessThanLowerThreshold comparison (those alarms never fired). Thread real alarm type into action history; honor ListMetrics RecentlyActive=PT3H. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…leak, tag-before-create Sign/Verify/GetPublicKey/DeriveSharedSecret/GenerateDataKeyPair(+WithoutPlaintext)/ GenerateMac/VerifyMac had no GrantTokens field — silently dropped, grant validity never enforced (disguised stub). Unrecognized KeySpec now 400 ValidationException (was 500). purgeKey drops leaked grantsByKey submap. Validate tags BEFORE creating key/replica (was leaving orphaned untagged keys); ReplicateKey bypassed validation. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…t, change-set exec gates, event pagination DeleteStack now idempotent no-op for missing stack (AWS models no not-found error). Block DeleteStack/UpdateStack that would drop an export still imported via Fn::ImportValue. ExecuteChangeSet gates on ExecutionStatus + deletes all stack change sets. Fix ChangeSetNotFound wire code (was ChangeSetNotFoundException). AUTO_EXPAND no longer satisfies IAM capability. DescribeStackEvents honors NextToken via pkgs/page. Emit UPDATE_FAILED event on template parse failure. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
… service deployments Restore never rebuilt serviceIndex (reconciler's only feed) — every service went invisible to deployment/scaling after restart. resourceTags side map absent from snapshot (tag data loss). DescribeServiceDeployments/List/Stop were disguised stubs (only test-seeded); now recorded by real CreateService/UpdateService/rollback + a map-key leak fixed on cluster/service delete. CreateCluster now honors capacityProviders/defaultStrategy/tags. Implement ContinueServiceDeployment. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
cloudwatch sweep (ede7169) changed PutMetricData to return only error (dropped the fabricated UnprocessedMetricData). Update the two cli.go composition-root call sites from `_, err :=` to `err :=`. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…fication CompleteLayerUpload missing RepositoryNotFound FK check + silently overwrote an already-registered layer (now LayerAlreadyExistsException). UploadLayerPart discarded partFirstByte — no sequencing check (now InvalidLayerPartException). PutImage trusted client imageDigest verbatim; now verified against manifest hash (ImageDigestDoesNotMatchException). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…; drop tracked junk DeleteDBInstance/DeleteDBCluster ignored SkipFinalSnapshot/FinalDBSnapshotIdentifier — never validated the combo or took a final snapshot (disguised stub). Added DeleteDBInstanceWithOptions/DeleteDBClusterWithOptions (additive — old signatures preserved for cloudformation callers). DescribeDBInstances now honors Filters. Remove stale tracked batch3_test.go.rej / batch3_test_pi.patch artifacts. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…on-stage move, idempotency Fix concurrent map write: List/Describe/GetResourcePolicy held RLock but the lazy *Store(region) helper writes b.secrets[region] on first touch. Rename IncludeDeleted->IncludePlannedDeletion and owned-by-me->owning-service (wrong wire keys, filters silently ignored). UpdateSecretVersionStage now requires RemoveFromVersionId to name the current holder. Add CreateSecret ClientRequestToken idempotency, NextRotationDate/SortBy. Route all timestamps through injectable clock. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
… + version selector Intelligent-Tiering now auto-upgrades to Advanced for >4KiB/policies (was hard reject). Parameter Policies require Advanced tier. Enforce 100-version cap (ParameterMaxVersionLimitExceeded, was silently evicting labeled versions) + fix parameterLabels leak. Enforce 15-level hierarchy limit. DescribeDocument no longer embeds full Content (real DocumentDescription has none). Resolve $DEFAULT vs $LATEST. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…issing wire fields Integration TimeoutInMillis ceiling/default was hardcoded 29000ms for all protocols — HTTP APIs allow up to 30000ms (valid values were rejected). Tag handlers now map ErrStageNotFound to 404 (was 500). Add absent wire fields: Integration TlsConfig + ConnectionType (default INTERNET + VPC_LINK validation), Stage ClientCertificateId + Tags (nested stage ARN), DomainName MutualTls + DomainNameArn. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
… coercion, remove/copy
All PATCH ops shared a flatten function handling only single-segment add/replace
with verbatim string copy — AWS always sends PatchOperation.Value as a JSON string,
so every bool/int field (tracingEnabled, minimumCompressionSize, apiKeyRequired)
failed to unmarshal, and multi-segment paths (/variables/{name}, /binaryMediaTypes,
/apiStages, throttle) were silently dropped — setting one stage variable never worked.
New patch.go with JSON-Pointer paths, value coercion, remove/copy. Fix
UpdateGatewayResponse full-replace-on-patch; add UpdateAccount CloudwatchRoleArn,
Stage cache-cluster fields.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…AND shards CreateStream(Tags)/AddTagsToStream/RemoveTags/ListTags used a parallel unpersisted Handler.tags map — tags vanished on restart. Routed all 6 ops through backend stream.Tags (persisted, single source). PutRecords on a missing stream returned 200 w/ per-record InternalFailure — now top-level ResourceNotFoundException; reject empty batch. ON_DEMAND streams get 4 shards (was 1, caller ShardCount ignored). DescribeStream shard pagination (Limit/ExclusiveStartShardId/HasMoreShards). Consumer 20-cap; persist account limit. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…RESS async, retry/catch runMapItem checked only the Go error, not res.Error — failing Map iterations were silently swallowed (nil holes, Map always succeeded). History recorder discarded all event detail (Input/Output/Resource/Error) — every event body was an empty shell. Allow async StartExecution on EXPRESS; StartSyncExecution on STANDARD now returns StateMachineTypeNotSupported. Separate Error/Cause in Catch. Add state-level Retry/Catch on Map+Parallel, Retry MaxDelaySeconds/JitterStrategy, Map ToleratedFailureCount/Percentage. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…sistence (Phase 3.3) containers map[region]map[name]*Container -> map[region]*store.Table[Container] (elasticache region-nested pattern; dynamic region set -> per-region Table.Snapshot, no fixed Registry). Direct/clean (Container.Name identity). mediastore had NO persistence -> built from scratch (Snapshot/Restore + StorageBackend iface + Handler delegation), n->y whole service. Added non-creating getContainer accessor to keep RLock paths safe (naive lazy-creator = concurrent-map-write hazard). paginationSecret unpersisted (sibling precedent). Net +33 backend. Snapshot version guard + full round-trip/version tests. -race + build green. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…e (Phase 3.3) jobs/resources/iterations -> store.Table (direct: JobID/Arn/FlywheelIterationID). tags/policies/policyRevisions raw (non-*T value maps, n->y). comprehend had NO persistence -> built from scratch (Snapshot/Restore + Handler delegation), n->y whole service (jobs/resources/iterations/tags/policies/policyRevisions). mu sync.RWMutex -> *lockmetrics.RWMutex (internal, matches templates). Net +19 backend. Snapshot version guard + full round-trip/version tests. Exported API unchanged; -race + build green. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…ase 3.3) lexicons/tasks -> store.Table (direct: Lexicon.Name/SpeechSynthesisTask.TaskID). tags raw (non-*T value map, n->y). voices static catalog untouched. polly had NO persistence -> built from scratch (Snapshot/Restore + Handler delegation), n->y whole service. Kept sync.RWMutex (polly never migrated to lockmetrics; out of mechanical scope). Net +9 backend. Snapshot version guard + full round-trip/version tests. Exported API unchanged; -race + build green. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
channels/originEndpoints/harvestJobs/packagingConfigurations -> store.Table (all direct, real ID). tags raw (non-*T value map, y->y). byChannel Index on originEndpoints (reverse-map pattern; replaces 2 linear scans in DeleteChannel cascade + ListOriginEndpoints filter; slices.Clone before cascade delete). Fixed dead-wiring: backend implemented Snapshot/Restore but Handler never delegated -> persistence never wired -> added Handler.Snapshot/Restore (dead->live). Net +114 backend. Snapshot version guard + full round-trip/version/index tests. Exported API unchanged; -race + build green. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
… (Phase 3.3) terminologies/parallelData/jobs -> store.Table (direct: Name/Name/JobID). tags raw (non-*T value map, n->y). translate had NO persistence -> built from scratch (Snapshot/Restore + Handler delegation), n->y whole service. Kept sync.RWMutex (never migrated to lockmetrics; out of mechanical scope). Net +43 backend. Snapshot version guard + full round-trip/version/handler-delegate tests. Exported API unchanged; -race + build green. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
transactions map[region]map[txID]*Transaction -> map[region]*store.Table (mediastore region-nested pattern; per-region txCounter means txID only region-unique, so region->Table not composite key). executedStatements (order- sensitive region-nested slice) + txCounter raw, y->y. Fixed dead-wiring: backend implemented Snapshot/Restore but Handler never delegated -> persistence never invoked -> added Handler.Snapshot/Restore (dead->live). Non-creating transactionRegion accessor for RLock safety. Net +101 backend. Snapshot version guard + full round-trip/version tests. Exported API unchanged; -race + build green. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…nce (Phase 3.3) resources -> store.Table (direct, TypeName|Identifier composite string key); requests -> store.Table (direct, RequestToken). clientTokens raw (non-*T value map, n->y). cloudcontrol had NO persistence -> built from scratch (Snapshot/Restore + Handler delegation), n->y whole service. ClientToken idempotency verified via round-trip post-restore. Net +5 backend. Snapshot version guard + full round-trip/ version tests. Exported API unchanged; -race + build green. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
monitors map[region]map[name]*Monitor -> flat store.Table (composite region|monitorName key + byRegion Index). Added Monitor.Region field (additive, wire-invisible; all responses go through DTOs). Probes untouched (nested []*Probe slice, not a map). arnIndex raw (n->n, write-only dead bookkeeping, left byte-for- byte). Fixed dead-wiring: backend implemented Snapshot/Restore but Handler never delegated -> persistence never registered -> added Handler.Snapshot/Restore (dead->live). slices.Clone before pagination sort (Index contract). Net +81 backend. Snapshot version guard + full round-trip/version tests. Exported API unchanged behaviorally; -race + build green. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…3.3, 0 convertible) connections map value connState embeds live downstream chan []byte -> leave-raw category (non-serializable live handle), 0/1 convertible (redshiftdata precedent). Added apigatewaymanagementapiSnapshotVersion=1 guard around the raw map only (version-mismatch/absent -> reset+nil). No dead-wiring bug (Handler already delegated). connections y->y (existing persistedConn DTO drops the channel). Net +94 (additive version field + guard + tests). Exported API unchanged; -race + build green. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…ce (Phase 3.3) resources map[kind]map[name]*Resource -> map[kind]*store.Table[Resource] (13 per-kind Tables, direct by Name; Name unique-within-kind matches prior per-kind map semantics). evaluations (order-sensitive slice values) + tags raw, n->y. arnIndex is cross-Table reverse index (can't be store.Index) -> rebuilt on Restore from each Resource ARN/Kind (dax rebuildTagIndex pattern). forecast had NO persistence -> built from scratch (Snapshot/Restore + Handler delegation), n->y whole service. Net -1 backend. Snapshot version guard + full round-trip/version tests. Exported API unchanged; -race + build green. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…ersistence (Phase 3.3) objects map[region]map[path]*Object -> map[region]*store.Table[Object] (mediastore region-nested pattern; per-region Table.Snapshot, no fixed Registry). Direct; added Object.Path field (additive, from normalizePath key) per hidden-ID gotcha. state()/stateRO() lazy/non-creating accessor split for RLock safety. ListItems ordering preserved (untouched sort.Slice by Name, asserted). mediastoredata had NO persistence -> built from scratch (Snapshot/Restore + Handler delegation), n->y whole service. Net +20 backend. Snapshot version guard + full round-trip/version/ ordering tests. Exported API unchanged; -race + build green. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
policies map[string]*storedPolicy -> store.Table (direct, PolicyID). counter scalar raw (y->y; maxCounterFromPolicies derives on legacy restore). ARN lookup kept as linear Range scan (no secondary index warranted). Fixed dead-wiring: backend implemented Snapshot/Restore but Handler never delegated -> dlm silently never persisted -> added Handler.Snapshot/Restore (dead->live). Net +120 backend. Snapshot version guard + full round-trip/version tests. Exported API unchanged; -race + build green. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…e 3.3, 0 convertible) dynamodbstreams owns zero local state: no backend.go, Handler.Streams is a StreamsBackend interface wired in cli.go directly to services/dynamodb's *InMemoryDB. Pure query/wire layer -> 0/0 maps, nothing to snapshot. Deliberately added NO persistence: a delegating Handler.Snapshot/Restore would double-register dynamodb's full snapshot under a second key + double-restore into the shared object (real behavior change, not a mechanical fix). All stream/record ordering lives in and is persisted by services/dynamodb, untouched. Added TestHandler_OwnsNoState guarding against accidental future double-registration. Test-only, no production change. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…d persistence (Phase 3.3) alternateContacts map[ContactType]*AlternateContact -> store.Table (direct, AlternateContactType). account had NO persistence -> built from scratch (Snapshot/Restore + StorageBackend iface + Handler delegation): contactInfo, regions, alternateContacts + scalars all n->y. NOTE: account has no provider.go / not in cli.go service list -> persistence not yet invoked at runtime (pre-existing registration gap, out of scope). Net +12 backend. Snapshot version guard + full round-trip/version tests. Exported API unchanged; -race + build green. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…ard (Phase 3.3) sessions/asyncInvocations -> store.Table (direct: Session.ID/AsyncInvocation. InferenceID). invocations []*Invocation raw (order-sensitive capped FIFO, no identity, y->y). nextID counter raw (y->y). Snapshot gained version guard (sagemakerruntimeSnapshotVersion=1; old no-version snapshots decode as 0 -> reset). No dead-wiring (Handler already delegated). evictOldest helper threaded idFn for Table.Delete (unexported, FIFO-by-CreatedAt semantics preserved). Net +94 backend. Full round-trip/version tests. Exported API unchanged; -race + build green. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…ckmetrics stragglers) Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…ression, needs sharding) Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…ollow-up) The eventbridge store conversion (b1a4815) changed InMemoryBackend.PutEvents from returning []EventResultEntry to ([]EventResultEntry, error), but two shared call sites in the wiring layer still called it as a bare statement -> compiled clean (build + vet green) but errcheck flagged the ignored error. Scoped per-service gates missed it: cli.go is package main at repo root, outside services/eventbridge/. - cli.go s3EventBridgeAdapter.PublishS3Event: void return (s3.EventBridgePublisher is fire-and-forget), so explicitly discard via _, _ = (errcheck-approved), behavior byte-for-byte preserved. - cli_adapters.go schedEventBusAdapter.PutSchedulerEvent: returns error, was unconditional 'return nil' -> propagate the PutEvents error (faithful: PutEvents could not error pre-3.3). Verified: golangci-lint v2.12.2 (repo config) on root package = 0 issues; go vet ./ clean. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…tack-nab) aws-sdk-go-v2 eks v1.88.x (dep upgrade f959827) added the CancelUpdate op; TestSDKCompleteness flagged it as neither supported nor acknowledged, failing deterministically. Added to the notImplemented slice per the tracked fix — the op genuinely isn't implemented yet (that's a parity task, not a datalayer one). eks was never part of the pkgs/store rollout; this is pre-existing SDK-upgrade fallout, not a store regression. eks -run TestSDKCompleteness + vet + lint green. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…00->30000) TestIntegration_APIGatewayV2_FullLifecycle creates an HTTP API but asserted the default TimeoutInMillis == 29000 (the WebSocket value). AWS defaults HTTP-API integration timeout to 30000ms; the backend returns 30000 via integrationTimeout MaxFor(ProtocolTypeHttp), correct since parity fix e069888. The test assertion predates that fix (last touched 11c24a8) and has been stale since Phase 2 — NOT a pkgs/store regression (apigatewayv2 store conversion 3a8aa26 did not touch the value). Fixed assertion to the AWS-correct 30000. go vet ./test/integration/ clean. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
APIConsoleMiddleware captured every request header verbatim into the console ring buffer (CapturedRequest.Headers), including Authorization, X-Amz-Security- Token, X-Amz-Credential/Signature, Cookie and API keys — those values then flowed to slog sinks, so secrets were stored + logged in clear text (CodeQL go/clear-text-logging, 26 HIGH alerts all tainted from this single source). Redact sensitive-header values ([REDACTED]) at the capture point, severing the taint to every downstream sink. Non-sensitive headers retained verbatim. Set is a setup-time local (allocated once per middleware construction; avoids a package global). Pre-existing issue (apiconsole.go predates Phase 3.3; not a store bug), surfaced by the branch CodeQL scan. Added TestAPIConsoleMiddleware_RedactsSensitive Headers. pkgs/logger build+vet+lint(0)+test green. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
… pages The frontend dep upgrade (deac816/f959827d: SvelteKit 2.69, Svelte 5.56, Vite 8) regressed the dashboard: every service page rendered a blank shell, failing ~42 e2e tests with Playwright locator timeouts (controls never mount). ui-lint, svelte-check and vitest all passed — the break was purely runtime. Root cause: under Vite 8's Rolldown bundler, default code-splitting hoisted shared AWS SDK/Smithy helpers into an arbitrary page-route chunk, then the shared SDK chunk imported them back -> circular import. That cycle left command-factory bindings (e.g. RDS/Neptune classBuilder) uninitialized when a route's top-level 'class extends factory(...)' ran, throwing 'TypeError: z is not a function' during hydration and blanking the whole app. Fix: manualChunks pins all node_modules @aws-sdk/@smithy code into one 'aws-sdk' chunk, so page nodes only import FROM it, never the reverse — breaking the cycle. Not a pkgs/store issue; pre-existing UI-upgrade fallout. Verified: full 'go test -tags=e2e ./test/e2e/...' PASS (313s); oxlint 0, svelte-check 0 errors, vite build clean. Built SPA output stays gitignored. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Collaborator
Author
📊 Code Coverage Report
Tip This project maintains a minimum coverage threshold of 85%. Maintain or improve coverage on new code to ensure long-term stability. Last updated: Sat, 11 Jul 2026 19:51:24 GMT |
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
1 participant
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.
No description provided.