> each worker gets only a shallow copy [4] of the process list. Now I think that is not the cause.
> the verify job run [3] showed no effect. As the tests use cli_inband API to execute CLI commands, I removed another cause of worker barrier, but patch set #4 of [2] still shows too few calls. Additionally, marking cli_inband as thread safe somehow reduced the dataplane performance (after the wait). Vratko. From: vpp-dev@lists.fd.io <vpp-dev@lists.fd.io> On Behalf Of Vratko Polak -X (vrpolak - PANTHEON TECHNOLOGIES at Cisco) via lists.fd.io Sent: Monday, 2022-June-20 19:26 To: vpp-dev@lists.fd.io Subject: [vpp-dev] wait CLI A CLI command "wait" was added in [0], and a related function vlib_process_suspend was there in the first open commit [1]. I guess both are intended to be used during initialization, forcing an exec script to slow down, e.g. for a DHCP response to arrive. But when I tried (in [2]) to teach "wait" it does not need to stop workers, the verify job run [3] showed no effect. If I am looking correctly, that is because each worker gets only a shallow copy [4] of the process list. Is this behavior intended? Is there a way to pause only CLI processing without also affecting worker threads? Vratko. [0] https://gerrit.fd.io/r/c/vpp/+/23004/10/src/vlib/unix/cli.c [1] https://gerrit.fd.io/r/c/vpp/+/23/3/vlib/vlib/node_funcs.h#326 [2] https://gerrit.fd.io/r/c/vpp/+/36444 [3] https://jenkins.fd.io/job/vpp-verify-master-ubuntu2004-x86_64/9215/ [4] https://gerrit.fd.io/r/c/vpp/+/23/3/vlib/vlib/threads.c#665
-=-=-=-=-=-=-=-=-=-=-=- Links: You receive all messages sent to this group. View/Reply Online (#21571): https://lists.fd.io/g/vpp-dev/message/21571 Mute This Topic: https://lists.fd.io/mt/91882341/21656 Group Owner: vpp-dev+ow...@lists.fd.io Unsubscribe: https://lists.fd.io/g/vpp-dev/leave/1480452/21656/631435203/xyzzy [arch...@mail-archive.com] -=-=-=-=-=-=-=-=-=-=-=-