go.mod: update to pgx v4.13

[rafiss]

fixes #62840

This also required an update to cockroach-go/v2 to get the crdbpgx
transaction retry helper.

The majority of changes here are because pgx now requires the context to
be passed in always.

Other changes:

• The way that pgx prepares statements has changed -- to disable
  automatic prepared statements, the PreferSimpleProtocol setting must be
  enabled when making the pgxpool.
• The noprepare setting was removed from workload. This functionality
  is not supported by pgx any more. pgx will either use prepared statements and
  cache them automatically or use the simple protocol.
• sqlproxy tests had to be modified since pgx now has more
  fallback logic to retry connections when using an sslmode of prefer
  or allow. pgx also now tries to resolve localhost as the IPv6 address
  :::1, which isn't well-supported by sqlproxy, so IP resolution was disabled
  for sqlproxy tests.
• CHANGEFEED tests had to be changed since pgx.Conn.Query now immediately
  tries fetching the results, which didn't work with the synchronization that was added
  to TestChangefeedDataTTL.
• COPY was fixed to correctly ignore CopyData after encountering an error.

Release note: None

[cockroach-teamcity]

This change is 

[otan]

some test failures but they look like you'd have to massage them differently, lgtm

[knz]

Reviewed 55 of 55 files at r1, all commit messages.
Reviewable status: complete! 0 of 0 LGTMs obtained (waiting on @adityamaru, @rafiss, and @stevendanna)

---

pkg/ccl/backupccl/backup_test.go, line 7297 at r1 (raw file):

				db, err := pgx.ConnectConfig(ctx, connCfg)
				assert.NoError(t, err)
				defer func() { _ = db.Close(ctx) }()

assert.NoError(t, db.Close(ctxo))

---

pkg/workload/kv/kv.go, line 444 at r1 (raw file):

		}
		defer func() {
			_ = tx.Rollback(ctx)

maybe assert the return value?

[rafiss]

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

---

pkg/ccl/backupccl/backup_test.go, line 7297 at r1 (raw file):

Previously, knz (kena) wrote…

assert.NoError(t, db.Close(ctxo))

done

---

pkg/workload/kv/kv.go, line 444 at r1 (raw file):

Previously, knz (kena) wrote…

maybe assert the return value?

done

[rafiss]

sqlproxy'sTestFailedConnection is failing in CI, but I cannot get it to fail even when running it under stress on a GCE Worker.

Will debug later today or tomorrow.

Update: This was failing because pgx now tries to resolve localhost to the IPv6 address :::1, and sqlproxy does not support IPv6. Updated the test to not use IPv6.

[rafiss]

@stevendanna if you have some time would you be able to help me with this?

I'm trying to upgrade our pgx dependency to use v4 across the board, and I've encountered an issue in TestChangefeedDataTTL which it seems like you have touched last.

The test is getting stuck at this line:

cockroach/pkg/ccl/changefeedccl/changefeed_test.go

Line 2328 in ab15e5f

dataExpiredRows := feed(t, f, "CREATE CHANGEFEED FOR TABLE foo")

The reason is because the sinklessFeed is gets stuck when making the changefeed:

cockroach/pkg/ccl/changefeedccl/testfeed_test.go

Line 186 in ab15e5f

c.rows, err = c.conn.Query(create, c.args...)

pgx v4 differs from the old pgx in that it immediately tries to fetch the result when running Query(). That means the synchronization logic in the test won't let the pgx.Conn.Query("CREATE CHANGEFEED ...") statement complete.

Do you have any other ideas on how to synchronize here while still achieving the goals of the test? The other option I thought of is to jump into a lower level and use jackc/pgconn to create the changefeed without trying to pull results from it right away, but that seems like a big refactor and isn't my first choice.

[rafiss]

I understand the test a bit better now so I tried to fix the test myself -- @stevendanna when you have a chance, please review the diff in pkg/ccl/changefeedccl/changefeed_test.go

[stevendanna]

Uff, apologies for the delay here. Thanks for digging into this. I think your fix looks reasonable. This test has been a thorn in our side for a bit as the existing coordination is a bit suspected. I noted a small nitpick but overall it

Reviewed 1 of 35 files at r3, 1 of 25 files at r9.
Reviewable status: complete! 1 of 0 LGTMs obtained (waiting on @adityamaru, @knz, and @rafiss)

---

pkg/ccl/changefeedccl/changefeed_test.go, line 2373 at r10 (raw file):

		resume <- struct{}{}
		dataExpiredRows := <-changefeedInit
		defer closeFeed(t, dataExpiredRows)

[nit] We could fail to clean up this feed if there are errors above in upsertRow() or forceTableGC.

---

pkg/ccl/changefeedccl/testfeed_test.go, line 129 at r10 (raw file):

}

var _ cdctest.TestFeed = (*sinklessFeed)(nil)

Thanks for adding these!

[rafiss]

Reviewable status: complete! 1 of 0 LGTMs obtained (waiting on @adityamaru, @knz, @rafiss, and @stevendanna)

---

pkg/ccl/changefeedccl/changefeed_test.go, line 2373 at r10 (raw file):

Previously, stevendanna (Steven Danna) wrote…

[nit] We could fail to clean up this feed if there are errors above in upsertRow() or forceTableGC.

in that case there would have already been a t.Fatal, or otherwise closeFeed itself will call t.Fatal

[rafiss]

Reviewable status: complete! 1 of 0 LGTMs obtained (waiting on @adityamaru, @knz, @rafiss, and @stevendanna)

---

pkg/ccl/changefeedccl/changefeed_test.go, line 2373 at r10 (raw file):

Previously, rafiss (Rafi Shamim) wrote…

in that case there would have already been a t.Fatal, or otherwise closeFeed itself will call t.Fatal

ah, i see your point. the execution could stop before reaching this defer. fixing it, nice catch!

[rafiss]

thanks for the reviews all!

bors r=otan,knz,stevendanna

[craig]

Build failed (retrying...):

• GitHub CI (Cockroach)

[craig]

Build succeeded:

• GitHub CI (Cockroach)