search

ClusterLabs / fence-agents

Fence agents
104 stars 160 forks source link

Question : When disable-timeout is enabled, the timeout behavior is different for each fence_agent. #431

Open HideoYamauchi opened 3 years ago

HideoYamauchi commented 3 years ago

Hi All,

If disable-timeout is enabled, you should get a Pacemaker timeout, but it actually works differently depending on the fence_agent. This may enable or disable the Pacemaker pcmk_xxxx_retries parameter, depending on the fence_agent. Originally, shouldn't all failures time out if disable-timeout is enabled?

Here are examples of fence_ipmitool and fence_vmware_rest, with the differences we're talking about in Bugzilla below.

I haven't checked all the fence_agents, but it seems that there are many fence_agents that make a difference.

Best Regards, Hideo Yamauchi.

oalbrigt commented 3 years ago

It should. So seems like we need to update it ro use -N/-R or combination of them to work optimally in this case: https://access.redhat.com/discussions/5474551

HideoYamauchi commented 3 years ago

Hi Oyvind,

Thanks!

Are other fence_agents okay? I think it is necessary to look comprehensively.

Best Regards, Hideo Yamauchi.

oalbrigt commented 3 years ago

If you run grep -e "run_command" -e "pexpect" agents/*/*.py you should get most of the agents that might have similar issues if the tool they use uses internal timeout.

HideoYamauchi commented 3 years ago

Hi Oyvind,

If you run grep -e "run_command" -e "pexpect" agents/*/*.py you should get most of the agents that might have similar issues if the tool they use uses internal timeout.

Okay!

Let's check all fence_agents once.

However, if there are a wide variety of external commands to be executed, it will be very troublesome to deal with and confirm. I think it is desirable to make it so that the timeout can be ignored only on the main body side of fence_agent, but I do not know if it is possible.

Best Regards, Hideo Yamauchi.