sql: add trace_id to show sessions

[maryliag]

Add the column trace_id to the result of SHOW SESSIONS.
Also adding the column to the views:
crdb_internal.node_sessions and crdb_internal.cluster_sessions.

Part Of #117625

Release note (sql change): Add column trace_id to the response of the SHOW SESSIONS command.

[cockroach-teamcity]

This change is 

[fqazi]

Reviewed 7 of 10 files at r1, 5 of 5 files at r2, all commit messages.
Reviewable status: complete! 1 of 0 LGTMs obtained (waiting on @xinhaoz)

[dhartunian]

just a nit and a question, not blocking.

Reviewed 7 of 10 files at r1, 5 of 5 files at r2, all commit messages.
Reviewable status: complete! 2 of 0 LGTMs obtained (waiting on @maryliag and @xinhaoz)

---

pkg/server/serverpb/status.proto line 1057 at r1 (raw file):

  uint32 pg_backend_pid = 19 [ (gogoproto.customname) = "PGBackendPID" ];

  // The ID the session's trace.

nit: ID of the session's trace

---

pkg/sql/conn_executor.go line 4163 at r2 (raw file):

		TotalActiveTime:            sessionActiveTime,
		PGBackendPID:               ex.planner.extendedEvalCtx.QueryCancelKey.GetPGBackendPID(),
		TraceID:                    uint64(ex.planner.extendedEvalCtx.Tracing.connSpan.TraceID()),

weird that we have to plumb so far to get the trace. there's no context that we could extract the trace from? that's how most of the other code works.

or does the caller already have a handle on a span that could be passed in?

[maryliag]

Reviewable status: complete! 0 of 0 LGTMs obtained (and 2 stale) (waiting on @dhartunian, @fqazi, and @xinhaoz)

---

pkg/server/serverpb/status.proto line 1057 at r1 (raw file):

Previously, dhartunian (David Hartunian) wrote…

nit: ID of the session's trace

Done

---

pkg/sql/conn_executor.go line 4163 at r2 (raw file):

Previously, dhartunian (David Hartunian) wrote…

weird that we have to plumb so far to get the trace. there's no context that we could extract the trace from? that's how most of the other code works.

or does the caller already have a handle on a span that could be passed in?

In this case we need to go this far. You can notice something similar happens on the line right above for the pg backend PID. The function I'm calling already handles the case when the tracing is on/off, is the span was closed, etc. It might be complicated to use something else and having to handle all of that again.

[dhartunian]

Reviewed 2 of 2 files at r3, all commit messages.
Reviewable status: complete! 0 of 0 LGTMs obtained (and 2 stale) (waiting on @maryliag and @xinhaoz)

---

pkg/ccl/logictestccl/testdata/logic_test/crdb_internal_tenant line 266 at r3 (raw file):

id  node_id  session_id  start  txn_string  application_name  num_stmts  num_retries  num_auto_retries  last_auto_retry_reason  isolation_level  priority  quality_of_service

query ITTTTTTTTTTTTTTTI colnames

Why does this new column have an I here? The pg_backend_pid is also an int, right? but here it's a T which I think implies Text.

[maryliag]

bors r+

[maryliag]

Reviewable status: complete! 0 of 0 LGTMs obtained (and 2 stale) (waiting on @dhartunian and @xinhaoz)

---

pkg/ccl/logictestccl/testdata/logic_test/crdb_internal_tenant line 266 at r3 (raw file):

Previously, dhartunian (David Hartunian) wrote…

Why does this new column have an I here? The pg_backend_pid is also an int, right? but here it's a T which I think implies Text.

I added an I since is an INT, the same way there is an I for the node_id, maybe the T on others were an oversight or the format itself didn't matter.
I have a following PR where I'm adding one more column, so I can use that and align all columns with the correct types.

[craig]

Build succeeded:

• Bazel Essential CI (Cockroach)