fix(spanner): fix session leak

[8398a7]

Fixed a problem where the session was not reused when a panic occurred.

Fixes #3460

[skuruppu]

Thanks for the fix @8398a7. I would like @olavloite to approve this one.

[olavloite]

Thanks, that's a good catch!

Would you mind changing the test case so that it does not hang indefinitely if the problem were to reappear? I've added a suggestion, but feel free to do it in another way if you have a better idea.

[olavloite]

The current test case will hang indefinitely if a session leak would be re-introduced by accident in the future. The below suggestion would cause the test to fail fast instead.

Suggested change

	_, err := client.ReadWriteTransaction(ctx, func(ctx context.Context, tx *ReadWriteTransaction) error {
	return nil
	})
	if err != nil {
	t.Fatalf("Unexpected error during transaction: %v", err)
	}
	if g, w := client.idleSessions.idleList.Len(), 1; g != w {
	t.Fatalf("idle session count mismatch.\nGot: %v\nWant: %v", g, w)
	}

[8398a7]

It's definitely a problem when it hangs.

Fixed.
df60414

[olavloite]

@tbpg This one also seems to be stuck waiting for kokoro to report the build status. Could it be that this is related to the fact that the PR is created from a fork? Other PRs do not seem to have this problem, AFAICT.

[tbpg]

Kokoro only starts automatically for a "trusted" group.

[olavloite]

Kokoro only starts automatically for a "trusted" group.

Ah. Thanks for finding that out. I guess that's reasonable.