Hi Julio,
If you unplug the cable that connects the controller to the database,
then there is nothing Sequoia can do since it is up to the PostgreSQL
driver to find out that the connection was broken and that the server is
dead. As the driver uses TCP connection, it will have to wait for the
TCP keepalive timeout to expire (so you have to tune your OS TCP stack
accordingly if you want a timely connection failure detection).
What you seem to describe seems to be a normal TCP connection behavior
under transient network failure. It is likely that the controller is
just waiting for the driver to complete the operation or report an error.
Note that the same behavior can happen with the group communication
between controllers if your network settings rely on TCP.
Hope this helps,
Emmanuel
I have a sequoia 2.10 installation against 2 back ends (postgresql 8.2)
In a big load one of the backends crashed and sequoia never noticed
it. I checked the status for them and always both of them were up and
running
After awhile the system got stuck.
Later we did another test were we just pull out the net work cable for
one of the back ends, wait for 5 minutes and the same behavior ,
sequoia reported that both back ends were up and running, then we
put back the cable and 2 minutes later everything went back to normal.
We did the same test with a low volume of transactions and sequoia
detected the failure.
Any ideas what's going on?
Thanks in advance .
JulioC.
_______________________________________________
Sequoia mailing list
[email protected]
https://forge.continuent.org/mailman/listinfo/sequoia
