fix(adbc): re-authenticate on Unauthenticated instead of failing permanently

[phillipleblanc]

Problem

SqlWithParams/QueryWithParams run over the ADBC FlightSQL connection, which authenticates only once, when it is opened: the Basic-auth handshake (OptionKeyUsername/OptionKeyPassword) yields a server-side session token that is then reused for every prepared statement on that long-lived connection.

If the server invalidates that session — most commonly when it expires after a period of inactivity — every subsequent prepared statement fails:

error preparing statement: Unauthenticated: [FlightSQL] Invalid credentials (Unauthenticated; Prepare)

SqlWithParams classified Unauthenticated as a backoff.Permanent error, so the connection never recovered on its own. The only workaround was to recreate the SpiceClient (i.e. restart the consuming process). The credentials are still valid — it is the cached session that is stale.

The plain (non-parameterized) Sql/Query path is unaffected because it re-runs AuthenticateBasicToken on every call, so it was already immune to session expiry.

Fix

• Detect ADBC auth failures via errors.As on adbc.Error (StatusUnauthenticated / StatusUnauthorized), with a message-substring fallback for cases where the typed error is not propagated.
• On an auth failure, re-open the ADBC connection (fresh handshake → new session) and retry the query once. A genuinely-bad credential therefore fails after one re-auth attempt rather than looping.
• Re-initialization is guarded by a new adbcMu mutex plus a staleness check (the connection a caller observed failing), so when many goroutines hit the expired session simultaneously the connection is re-opened at most once and the rest reuse it.
• Refactored the existing backoff/transient-retry logic into execADBCWithBackoff (no behavior change for non-auth errors).

Tests

• New unit test TestIsADBCAuthError (no server required) covering typed, wrapped, fallback, and non-auth cases.
• go build, go vet, gofmt -l, and go test -race on the new logic all pass.
• The only failing tests locally are the pre-existing local-runtime integration tests (taxi_trips dataset / :8090 runtime) that require the CI spiced harness — they fail identically on clean trunk, unrelated to this change.

Release / downstream

Needs a tagged release (e.g. v8.0.2) so consumers can pick it up; the immediate motivation is a downstream service whose long-lived control-plane ADBC connection went idle past the server's session TTL and then failed every parameterized query until restart.

[Copilot]

Pull request overview

This PR improves resiliency of the parameterized ADBC FlightSQL query path (SqlWithParams/QueryWithParams) by detecting authentication/session-expiry failures, re-opening the long-lived ADBC connection, and retrying the query once so callers don’t need to recreate the SpiceClient.

Changes:

• Add auth-failure detection (isADBCAuthError) and retry-once reauthentication behavior for SqlWithParams.
• Refactor the transient retry/backoff logic into execADBCWithBackoff (intended to preserve existing behavior for non-auth errors).
• Add a unit test for auth-error classification (TestIsADBCAuthError) and introduce an adbcMu mutex to coordinate reinitialization.

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 4 comments.

File	Description
client.go	Adds an adbcMu mutex field intended to coordinate ADBC connection reinitialization.
adbc.go	Implements auth-error detection, reinit-on-auth-failure retry behavior, and refactors ADBC backoff logic.
adbc_reauth_test.go	Adds unit tests for isADBCAuthError behavior.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.