sql: implement idle_in_transaction_session_timeout

[RichardJCai]

Release note (sql change): Implement IdleInTransactionSessionTimeout
variable to allow terminating sessions that are idle in a transaction
past the provides threshold.

Set the variable by using SET idle_in_transaction_session_timeout = 'time'.

Sessions that are idle in OPEN, ABORTED and DONE(COMMITWAIT) transaction states
will be terminated if the user idles longer than the threshold time.

Fixes #5924

[cockroach-teamcity]

This change is 

[solongordon]

I just tested this out manually and something seems off. I set the timeout variable and then ran BEGIN; followed immediately by SELECT 1. Then no matter how long I wait it doesn't seem like the session is canceled. Could you take a look?

Reviewable status: complete! 0 of 0 LGTMs obtained (waiting on @knz and @solongordon)

[solongordon]

I take it back, I lost my session setting without realizing it.

Reviewable status: complete! 0 of 0 LGTMs obtained (waiting on @knz)

[solongordon]

My only real complaint is the same as last time, which is that the tests take ~30 seconds to run. Any chance you could slim that down a bit without introducing flakiness?

Reviewable status: complete! 1 of 0 LGTMs obtained (waiting on @knz and @RichardJCai)

---

pkg/sql/conn_executor_exec.go, line 153 at r1 (raw file):

		case stateOpen:
			// Only start timeout if the statement is executed in an
			// implicit transaction.

s/implicit/explicit/

---

pkg/sql/run_control_test.go, line 514 at r1 (raw file):

	time.Sleep(3 * time.Second)
	// The connection should still be alive.
	err = conn.PingContext(ctx)

Isn't this the same as what you test at the beginning of TestIdleInTransactionSessionTimeout?

[RichardJCai]

My only real complaint is the same as last time, which is that the tests take ~30 seconds to run. Any chance you could slim that down a bit without introducing flakiness?

I'm going to try to decrease the time, can increase it if it introduces flakiness on CI.

[RichardJCai]

Reviewable status: complete! 1 of 0 LGTMs obtained (waiting on @knz, @RichardJCai, and @solongordon)

---

pkg/sql/run_control_test.go, line 514 at r1 (raw file):

Previously, solongordon (Solon) wrote…

Isn't this the same as what you test at the beginning of TestIdleInTransactionSessionTimeout?

Yeah I think this is unnecessary

[RichardJCai]

bors r+

[craig]

Build succeeded:

• GitHub CI (Cockroach)