rpc: Handle closed error #100525
rpc: Handle closed error #100525
Conversation
|
This change is |
We close the listener before closing the connection. This can result in a spurious failure due to the Listener also closing our connection. Epic: none Fixes: cockroachdb#100391 Fixes: cockroachdb#77754 Informs: cockroachdb#80034 Release note: None
andrewbaptist
force-pushed
the
2023-04-03-handle-ErrClosed
branch
from
d7cb57f to
41288e1
Compare
|
I fixed the first issue ( I can run the test for >30min without failure with both of these fixes. |
| // TODO(baptist): Better understand when this happens. It appears we can get | ||
| // spurious connections to other tests on a stress run. This has been | ||
| // happening for a while, but only comes out rarely when this package is | ||
| // stressed. This test is very aggressive since it is calling GRPCDialNode in | ||
| // a busy loop for 50ms. |
There was a problem hiding this comment.
I think this is a known problem. Test servers bind to port 0 to let the OS pick a random unused port, but as soon as we close it then a different test can be randomly assigned that very same port number, so if we later try to connect to that port we get unexpected results. This typically causes problems when we attempt to restart a stopped server reusing the same port number it was previously assigned, but someone else bound to it in the meanwhile.
andrewbaptist
added
the
backport-23.1.x
|
bors r=erikgrinaker |
|
Build succeeded: |
We close the listener before closing the connection. This can result in a spurious failure due to the Listener also closing our connection.
Epic: none
Fixes: #100391
Fixes: #77754
Informs: #80034
Release note: None