arc: add output plugin for Arc columnar database

[mmatczuk]

No description provided.

[claude]

Commits
LGTM

Review
Clean new component implementation. The arc output follows project conventions for registration, config spec, bundle imports, public wrapper, info.csv metadata, license headers, error handling, and context propagation. One naming convention violation in field name constants.

1. Field name constants lack component prefix (internal/impl/arc/output.go#L47-L57): Constants like fieldToken, fieldDatabase, etc. should follow the project convention <componentAbbrev>Field<Name> (e.g., aoFieldToken, aoFieldDatabase), as documented in the project Go patterns and used consistently across all other components.

[claude]

Bug: rowCount and err in the require.Eventually format args are evaluated at call time — before the polling loop runs — so they will always be 0 and nil (their zero values). On failure the message will read "expected N rows, got 0 (last err: <nil>)" regardless of what actually happened, making CI failures hard to debug.

Fix: format the message inside the condition closure instead. For example, capture a lastMsg string variable and update it each iteration, then use require.Eventuallyf or move the assertion outside the Eventually and check the final values after the loop.

[claude]

Commits
LGTM

Review
New arc output component for writing to an Arc columnar database via msgpack ingestion. Implementation follows project patterns: proper init() registration with MustRegisterBatchOutput, field name constants with ao prefix, ParsedConfig extraction, *service.Resources logger pattern, Apache 2.0 headers, public wrapper, community + cloud bundle imports, and info.csv entry. Unit and integration tests cover columnar/row formats, compression, auth, error handling, and timestamp conversion.

1. Integration test requireArcQueryRowCount produces misleading error messages on failure (integration_test.go#L106-L112): rowCount and err passed as require.Eventually format args are captured at call time (zero values), not when the assertion fails. On timeout the message will always say "expected N rows, got 0 (last err: <nil>)" regardless of the actual last observed state.

[claude]

Commits
LGTM

Review
New arc output component for writing to an Arc columnar database via MessagePack ingestion. Adds implementation, unit tests, integration tests, public wrapper, bundle registration (community + cloud), info.csv metadata, and autogenerated docs. Code follows project patterns correctly: MustRegisterBatchOutput in init(), field name constants with ao prefix, proper error wrapping with gerund form and %w, context propagation through component methods, safe sync.Pool usage for compression writers, header injection prevention on token and database name, and idempotent Close.

LGTM

[xe-nvdk]

Why doing this? Hard to track contributors, doesn't feel like a good practice. Can you clarify?

[mmatczuk]

@xe-nvdk to streamline things on your end like failing CI and to be able to include it in a release. Cheers.

[josephwoodward]

Should we be concerned with map's random ordering here?

[josephwoodward]

Should we document/log that any message field named "time" is dropped silently?

[josephwoodward]

Given we return on err here, would it not be better returning earlier (within the case)?

[josephwoodward]

suggestion:

var (
    payload any
    err error
)

[josephwoodward]

Not sure how helpful this error message prefix is (message %d), Is there a clearer prefix we could use?

[josephwoodward]

None of these comments are blockers and can be addressed post merge.