19583b7229
0069-containerd-add-check-in-spec.patch 0070-containerd-kill-container-init-process-if-runc-start.patch 0071-containerd-fix-containerd-shim-residual-when-kill-co.patch 0072-containerd-fix-deadlock-on-commit-error.patch 0073-containerd-backport-upstream-patches.patch 0074-containerd-fix-exec-event-missing-due-to-pid-reuse.patch 0075-containerd-fix-dm-left-when-pause-contaienr-and-kill-shim.patch 0076-containerd-fix-start-container-failed-with-id-exists.patch 0077-containerd-drop-opt-package.patch 0078-containerd-bump-containerd-ttrpc-699c4e40d1.patch 0079-containerd-fix-race-access-for-mobySubcribed.patch 0080-containerd-improve-log-for-debugging.patch 0081-containerd-reduce-permissions-for-bundle-di.patch 0082-containerd-fix-publish-command-wait-block-for.patch 0083-containerd-optimize-cgo-compile-options.patch Signed-off-by:songyanting <songyanting@huawei.com>
72 lines
3.3 KiB
Diff
72 lines
3.3 KiB
Diff
From dded5a0253fbfd3c75c6d73a890049c832374545 Mon Sep 17 00:00:00 2001
|
|
From: jingrui <jingrui@huawei.com>
|
|
Date: Sat, 20 Feb 2021 09:06:22 +0800
|
|
Subject: [PATCH] containerd: fix exec event missing due to pid reuse
|
|
|
|
When many exec request exit at nearly sametime, the Exit can match with
|
|
wrong process and return directly, the event for right process will lost
|
|
in this case.
|
|
|
|
time="2021-02-19T21:10:12.250841280+08:00" level=info msg=event Pid=11623 containerID=a32a1b7923db55ebdc7483e2b9cd986e5efc750b989ad3507eb866835e8e37f4 execID=0b412ecaed98f9ea71168599a9363b8aa3b047187eadaa74973bb6c63a66118d module=libcontainerd namespace=moby topic=/tasks/exec-started
|
|
time="2021-02-19T21:10:12+08:00" level=info msg="try publish event(1) /tasks/exit &TaskExit{ContainerID:a32a1b7923db55ebdc7483e2b9cd986e5efc750b989ad3507eb866835e8e37f4,ID:0b412ecaed98f9ea71168599a9363b8aa3b047187eadaa74973bb6c63a66118d,Pid:11623,ExitStatus:0,ExitedAt:2021-02-19 21:10:12.27697416 +0800 CST m=+1893.164673481,} <nil>"
|
|
time="2021-02-19T21:11:02.944643980+08:00" level=debug msg="starting exec command 64cd335311e9b3c1c11e7360a374e3218efeb02e6578d7bc0811bad3f1820e16 in container a32a1b7923db55ebdc7483e2b9cd986e5efc750b989ad3507eb866835e8e37f4"
|
|
time="2021-02-19T21:11:06.201162360+08:00" level=debug msg="event published" ns=moby topic="/tasks/exec-started" type=containerd.events.TaskExecStarted
|
|
time="2021-02-19T21:11:57.961615320+08:00" level=warning msg="Ignoring Exit Event, no such exec command found" container=a32a1b7923db55ebdc7483e2b9cd986e5efc750b989ad3507eb866835e8e37f4 exec-id=0b412ecaed98f9ea71168599a9363b8aa3b047187eadaa74973bb6c63a66118d exec-pid=11623
|
|
|
|
From logs above, execID=0b412ecae with Pid=11623 exit and event
|
|
published, but new exec execID=64cd335 command reuse the Pid, but Exit
|
|
event still match previous execID=0b412ecae. so exit event for
|
|
execID=64cd335 will lost.
|
|
|
|
Change-Id: If591a282a1cc0305758130a936ee8b92c88acc6c
|
|
Signed-off-by: jingrui <jingrui@huawei.com>
|
|
---
|
|
runtime/v1/linux/proc/exec.go | 4 ++++
|
|
runtime/v1/shim/service.go | 6 +++++-
|
|
2 files changed, 9 insertions(+), 1 deletion(-)
|
|
|
|
diff --git a/runtime/v1/linux/proc/exec.go b/runtime/v1/linux/proc/exec.go
|
|
index ea40cb5b8..a5f40bd63 100644
|
|
--- a/runtime/v1/linux/proc/exec.go
|
|
+++ b/runtime/v1/linux/proc/exec.go
|
|
@@ -86,6 +86,10 @@ func (e *execProcess) ExitedAt() time.Time {
|
|
}
|
|
|
|
func (e *execProcess) SetExited(status int) {
|
|
+ e.pid.Lock()
|
|
+ e.pid.pid = -1
|
|
+ e.pid.Unlock()
|
|
+
|
|
e.mu.Lock()
|
|
defer e.mu.Unlock()
|
|
|
|
diff --git a/runtime/v1/shim/service.go b/runtime/v1/shim/service.go
|
|
index 7e07ab011..7d7327cd8 100644
|
|
--- a/runtime/v1/shim/service.go
|
|
+++ b/runtime/v1/shim/service.go
|
|
@@ -548,8 +548,13 @@ func (s *Service) checkProcesses(e runc.Exit) {
|
|
log.G(s.context).WithError(err).Error("failed to check shouldKillAll")
|
|
}
|
|
|
|
+ match := 0
|
|
for _, p := range s.processes {
|
|
if p.Pid() == e.Pid {
|
|
+ match++
|
|
+ if match > 1 {
|
|
+ logrus.Warnf("exit for pid=%d match %d processes", e.Pid, match)
|
|
+ }
|
|
if ip, ok := p.(*proc.Init); ok {
|
|
ns := filepath.Base(filepath.Dir(ip.Bundle))
|
|
events.ExitAddFile(ns, events.ExitFile(s.id, uint32(e.Pid), uint32(e.Status)), "init exited")
|
|
@@ -591,7 +596,6 @@ func (s *Service) checkProcesses(e runc.Exit) {
|
|
ExitStatus: uint32(e.Status),
|
|
ExitedAt: p.ExitedAt(),
|
|
}
|
|
- return
|
|
}
|
|
}
|
|
}
|
|
--
|
|
2.17.1
|
|
|