[coverage] Conformance findings: SESSION-017

[peco-engineer-bot]

Summary

Surfaced by the multi-language coverage fan-out while conformance-testing these SPEC-IDs against databricks/databricks-sql-go. Each finding is committed as an expected-failure (xfail) test in the coverage PR — the test asserts the CORRECT (post-fix) behavior and stays red until THIS driver (databricks/databricks-sql-go) is fixed, then flips green as a tripwire.

Findings

• SESSION-017 [thrift]: Ping (driver.Pinger) on a CLOSED connection still probes the server: it submits select 1 on the disposed session handle (rejected INVALID_STATE: Invalid SessionHandle) instead of the spec's zero round trips, so post-close validity polling costs a doomed round-trip per poll (PECOBLR-2093)
  • failing test: TestActiveConnectionValidationProbesTheServer (see the coverage PR diff under tests/)
• SESSION-017: Ping (driver.Pinger) has no local closed-session guard: a connection-validity check on a CLOSED connection still submits the select 1 test query on the disposed session handle (server rejects with INVALID_STATE: Invalid SessionHandle), so validity polling after close costs a doomed server round-trip per poll instead of the spec's zero round trips; the thrift backend is affected, the kernel/SEA backend correctly short-circuits locally (PECOBLR-2093 / PECOBLR-2197)

Reproduce & Expected

SESSION-017 — Validates the ACTIVE side of connection validation: when active connection testing is ENABLED, the driver's "is this connection still usable?" check performs a real, lightweight server round-trip (a…

Reproduce:

• Fresh, reachable connection with active connection validation enabled
• Baseline before the check, so session-open traffic is not miscounted as
  the probe.
• Ask the driver whether the connection is still usable, via its own entry
  point (SQLGetConnectAttr(SQL_ATTR_CONNECTION_DEAD) / isValid(30) /
  PingContext). With active validation enabled this must round-trip the
  SELECT 1 test query to the server.
• Close the connection (CloseSession / DeleteSession)
• Baseline again, so the close traffic is not counted below
• The same check on the now-closed connection

Expected (per the shared spec):

• completes without an exception
• [thrift] exactly 0 ExecuteStatement call(s)
• [sea] exactly 0 ExecuteStatement call(s)
• full assertion contract:

result:
- connection_reported_alive: true
  label: while_live
  description: "A fresh, reachable connection is reported ALIVE even though the check\n\
    actively probed the server \u2014 the probe succeeded.\n"
- connection_reported_dead: true
  label: after_close
  description: 'After close the same check reports DEAD/invalid. The DELIVERY of that

    answer is surface-specific and not constrained here: a returned flag

    (ODBC SQL_CD_TRUE, JDBC isValid() == false) and a raised

    closed/invalid-handle error (Go''s Ping on a closed conn) both satisfy it.

    The contract is "not reported alive", not the mechanism.

    '
- no_exception: true
  label: while_live
  description: On a live connection the validity check itself never throws.
protocol:
  thrift:
  - call_min:
      method: ExecuteStatement
      min: 1
      label: while_live
  - call_count:
      method: ExecuteStatement
      expected: 0
      label: after_close
  sea:
  - call_min:
      operation: ExecuteStatement
      min: 1
      label: while_live
  - call_count:
      operation: ExecuteStatement
      expected: 0
      label: after_close

Context

• The behavior was first fixed in a DIFFERENT driver — reference PR: https://github.com/databricks/databricks-odbc/pull/299 — which seeded the shared language-neutral spec. This issue tracks the same conformance gap in databricks/databricks-sql-go; the reference PR is for cross-referencing the intended behavior, NOT a change to this repo.
• Coverage PR carrying the reproducing xfail test(s): https://github.com/databricks/databricks-driver-test/pull/1360