[steiler]

Ticket 06 of the #460 + #471 PR-split reorg
(see graph.md/index.md in the tracking gist for the full stack). Base:
cache-interface-prep (#495).

Scope

Read-only ConfigServerCache (Cache.Type: config-server): local-read seam,
fake for tests, real gRPC client, and wiring into Server.createCacheClient.
No writes — IntentWriter stays a no-op (noopIntentWriter, composed via
NewConfigServerClient) until the next ticket in the stack.

Cross-repo dependency

Depends on sdcio/sdc-protos#134, a fresh rebase of the dangling
config-server-cache-backend branch's config_read.proto/ConfigReadService
commit onto the tip of sdcio/sdc-protos#123 (the sensitive-flag rebase
sdcio/data-server#493 already pinned to). go.mod is pinned to that
commit; swap for a tagged sdc-protos release once #123/#134 merge.

What shipped

• configserver.LocalConfigReader seam (Get/List) + FakeLocalConfigReader
  for tests, Document/Target/ConfigBlob types per the ADR's field-mapping
  table.
• ConfigServerCache (pkg/cache/configserver.go): real-intent reads go
  through the seam and wrap into importer.IntentAdapter; writes are
  unconditional no-ops; running is served from its own independent
  in-memory, per-instance store, entirely separate from the seam.
• GRPCConfigReader: real seam implementation over the generated
  config_read.ConfigReadServiceClient, dialed by
  Server.createConfigServerCacheClient.
• Cache.Type: config-server now requires only Address (Namespace config
  field deleted) — ConfigServerCache derives the per-call target
  namespace/name from the datastore name itself (splitDatastoreName),
  correctly serving datastores across multiple Kubernetes namespaces from one
  process.
• Intent names are emitted GVKNSN-style (<namespace>.<name>), matching how
  config-server keys TransactionSet owners, with a lookup-side strip
  (lookupConfigName) so Get/Exists still work off either form.
• mergeConfigBlobs now handles config-server's root-path ("/") blob shape
  used by LoadAllButRunningIntents/revert-after-sync.

Adapted from the source branch

• Re-derived against this ticket's actual, narrower IntentAdapter/
  ImportConfigAdapter split and noopIntentWriter composition (both from
  the prior cache-interface-prep ticket), rather than the single
  ImportConfigAdapter shape / concrete-ConfigServerCache-implements-
  `IntentWriter shape some of the original commits were written against.
• Per the prior ticket's follow-up note: InstanceCreate seeds running
  with an empty &tree_persist.Intent{} (not nil) from the start, and
  InstanceRunningGet has no ErrRunningNotFound/nil-guard — built in
  correctly rather than reintroduced-then-fixed.
• Dropped two late "rename"/"colocate" commits from the source branch
  (706ecb6, 2e8ffc6): both rename ConfigReadService → ConfigSnapshotService
  and touch a pkg/cache/configsnapshot/ package that belongs to the next
  ticket's write-backend work (ConfigSnapshotService gains Modify/Delete
  there) — premature here, where the service is still read-only
  ConfigReadService.
• Dropped 8bebc69 (GetIntent redaction-scope doc comment): stale relative to
  the already-shipped always-union hardening from the sensitive-redaction
  ticket; the current comments already describe that behavior correctly.
• Dropped .scratch/ planning artifacts and doc-only "mark ticket done"
  commits, consistent with prior tickets in this stack.

Verification

go build ./..., go vet ./..., go test ./... (full repo), and
golangci-lint run ./... all pass.

[alexandernorth]

lgtm, one minor suggestion

[codecov]

Codecov Report

❌ Patch coverage is 88.31615% with 34 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
pkg/cache/configserver.go	88.18%	9 Missing and 4 partials ⚠️
pkg/server/cache.go	50.00%	7 Missing and 1 partial ⚠️
pkg/cache/configserver/merge.go	90.16%	3 Missing and 3 partials ⚠️
pkg/cache/configserver/grpc.go	88.37%	5 Missing ⚠️
pkg/cache/configserver/fake.go	94.28%	1 Missing and 1 partial ⚠️

📢 Thoughts on this report? Let us know!

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[faebr]

is it possible that here the values can be lost in case a path in the config ends with a list entry?

Example config:
`apiVersion: config.sdcio.dev/v1alpha1
kind: Config
metadata:
name: intf-uplink
namespace: default
labels:
config.sdcio.dev/targetName: dev1
config.sdcio.dev/targetNamespace: default
spec:
priority: 10
config:

• path: /interface[name=ethernet-1/1]
  value:
  admin-state: enable
  description: uplink`

And what data server builds from it:

{"interface":[{"name":"ethernet-1/1"}]}

Config-server itself still sends the full intent with TransactionSet, so the first apply to the device is correct. The lost leaves only affect intents that data-server reads back through this backend:

Next change to the intent: the old version data-server reads back is missing those leaves, so leaves that were removed in the new spec are never deleted from the device.
Rollback: the restored intent is missing those leaves.
BlameConfig / GetIntent: the leaves have no owner.
Deviations: the leaves look like running values that no intent manages ("unhandled").
Changes to other intents: a transaction for another intent on the same device rebuilds the tree from the cache. Leaves only this intent set lose their owner, and may be treated as deleted or no longer managed.

[faebr]

As far as I understood config-server will never send that type of config, so the path will only be / so up to you if you see this as an issue for the future or not