* [PATCH] cgroup: add tracepoints to track cgroup events
@ 2014-08-21 3:46 ` Andrea Righi
0 siblings, 0 replies; 13+ messages in thread
From: Andrea Righi @ 2014-08-21 3:46 UTC (permalink / raw)
To: Tejun Heo, Li Zefan, Steven Rostedt, Ingo Molnar
Cc: cgroups, linux-kernel, Andrea Righi, Matt Heaton
This patch adds the following tracepoints:
o trace_cgroup_create when a new cgroup is created
o trace_cgroup_destroy when a cgroup is removed
o trace_cgroup_task_migrate when a task/thread is moved from a cgroup to another
The purpose of these tracepoints is to identify and help cgroup "managers" to
diagnose problems and detect when they are doing an excessive amount of work.
Signed-off-by: Matt Heaton <matt@betterlinux.com>
Signed-off-by: Andrea Righi <andrea@betterlinux.com>
---
include/trace/events/cgroup.h | 95 +++++++++++++++++++++++++++++++++++++++++++
kernel/cgroup.c | 14 ++++++-
2 files changed, 108 insertions(+), 1 deletion(-)
create mode 100644 include/trace/events/cgroup.h
diff --git a/include/trace/events/cgroup.h b/include/trace/events/cgroup.h
new file mode 100644
index 0000000..937b41e
--- /dev/null
+++ b/include/trace/events/cgroup.h
@@ -0,0 +1,95 @@
+#undef TRACE_SYSTEM
+#define TRACE_SYSTEM cgroup
+
+#if !defined(_TRACE_CGROUP_H) || defined(TRACE_HEADER_MULTI_READ)
+#define _TRACE_CGROUP_H
+
+#include <linux/cgroup.h>
+#include <linux/tracepoint.h>
+
+#define TRACE_CGROUP_PATH_MAX 256
+
+#ifdef CREATE_TRACE_POINTS
+static inline void cgroup_safe_path(struct cgroup *cgrp, char *buf,
+ size_t buflen)
+{
+ char *path = cgroup_path(cgrp, buf, buflen);
+ size_t len;
+
+ if (likely(path)) {
+ /* NOTE: path is always NULL terminated */
+ len = strlen(path);
+ memmove(buf, path, len);
+ buf[len] = '\0';
+ } else {
+ strncpy(buf, "(NULL)", buflen);
+ }
+}
+#endif
+
+TRACE_EVENT(cgroup_create,
+
+ TP_PROTO(struct cgroup *cgrp),
+
+ TP_ARGS(cgrp),
+
+ TP_STRUCT__entry(
+ __array(char, name, TRACE_CGROUP_PATH_MAX)
+ ),
+
+ TP_fast_assign(
+ cgroup_safe_path(cgrp, __entry->name, TRACE_CGROUP_PATH_MAX);
+ ),
+
+ TP_printk("%s", __entry->name)
+);
+
+TRACE_EVENT(cgroup_destroy,
+
+ TP_PROTO(struct cgroup *cgrp),
+
+ TP_ARGS(cgrp),
+
+ TP_STRUCT__entry(
+ __array(char, name, TRACE_CGROUP_PATH_MAX)
+ ),
+
+ TP_fast_assign(
+ cgroup_safe_path(cgrp, __entry->name, TRACE_CGROUP_PATH_MAX);
+ ),
+
+ TP_printk("%s", __entry->name)
+);
+
+TRACE_EVENT(cgroup_task_migrate,
+
+ TP_PROTO(struct cgroup *old_cgrp, struct cgroup *new_cgrp,
+ const struct task_struct *p),
+
+ TP_ARGS(old_cgrp, new_cgrp, p),
+
+ TP_STRUCT__entry(
+ __field(pid_t, pid)
+ __array(char, old_name, TRACE_CGROUP_PATH_MAX)
+ __array(char, new_name, TRACE_CGROUP_PATH_MAX)
+ __array(char, comm, TASK_COMM_LEN)
+ ),
+
+ TP_fast_assign(
+ __entry->pid = p->pid;
+ memcpy(__entry->comm, p->comm, TASK_COMM_LEN);
+ cgroup_safe_path(old_cgrp, __entry->old_name,
+ TRACE_CGROUP_PATH_MAX);
+ cgroup_safe_path(new_cgrp, __entry->new_name,
+ TRACE_CGROUP_PATH_MAX);
+ ),
+
+ TP_printk("pid=%d comm=%s from=%s to=%s",
+ __entry->pid, __entry->comm,
+ __entry->old_name, __entry->new_name)
+);
+
+#endif /* _TRACE_CGROUP_H */
+
+/* This part must be outside protection */
+#include <trace/define_trace.h>
diff --git a/kernel/cgroup.c b/kernel/cgroup.c
index 7dc8788..00a50b9 100644
--- a/kernel/cgroup.c
+++ b/kernel/cgroup.c
@@ -60,6 +60,9 @@
#include <linux/atomic.h>
+#define CREATE_TRACE_POINTS
+#include <trace/events/cgroup.h>
+
/*
* pidlists linger the following amount before being destroyed. The goal
* is avoiding frequent destruction in the middle of consecutive read calls
@@ -2014,6 +2017,7 @@ struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset)
* Must be called with cgroup_mutex, threadgroup and css_set_rwsem locked.
*/
static void cgroup_task_migrate(struct cgroup *old_cgrp,
+ struct cgroup *new_cgrp,
struct task_struct *tsk,
struct css_set *new_cset)
{
@@ -2022,6 +2026,8 @@ static void cgroup_task_migrate(struct cgroup *old_cgrp,
lockdep_assert_held(&cgroup_mutex);
lockdep_assert_held(&css_set_rwsem);
+ trace_cgroup_task_migrate(old_cgrp, new_cgrp, tsk);
+
/*
* We are synchronized through threadgroup_lock() against PF_EXITING
* setting such that we can't race against cgroup_exit() changing the
@@ -2274,7 +2280,7 @@ static int cgroup_migrate(struct cgroup *cgrp, struct task_struct *leader,
down_write(&css_set_rwsem);
list_for_each_entry(cset, &tset.src_csets, mg_node) {
list_for_each_entry_safe(task, tmp_task, &cset->mg_tasks, cg_list)
- cgroup_task_migrate(cset->mg_src_cgrp, task,
+ cgroup_task_migrate(cset->mg_src_cgrp, cgrp, task,
cset->mg_dst_cset);
}
up_write(&css_set_rwsem);
@@ -2988,6 +2994,7 @@ static int cgroup_rename(struct kernfs_node *kn, struct kernfs_node *new_parent,
if (cgroup_on_dfl(cgrp))
return -EPERM;
+ trace_cgroup_destroy(cgrp);
/*
* We're gonna grab cgroup_mutex which nests outside kernfs
* active_ref. kernfs_rename() doesn't require active_ref
@@ -3004,6 +3011,9 @@ static int cgroup_rename(struct kernfs_node *kn, struct kernfs_node *new_parent,
kernfs_unbreak_active_protection(kn);
kernfs_unbreak_active_protection(new_parent);
+
+ trace_cgroup_create(cgrp);
+
return ret;
}
@@ -4587,6 +4597,7 @@ static int cgroup_mkdir(struct kernfs_node *parent_kn, const char *name,
goto out_free_id;
}
cgrp->kn = kn;
+ trace_cgroup_create(cgrp);
/*
* This extra ref will be put in cgroup_free_fn() and guarantees
@@ -4791,6 +4802,7 @@ static int cgroup_destroy_locked(struct cgroup *cgrp)
list_del_init(&cgrp->release_list);
raw_spin_unlock(&release_list_lock);
+ trace_cgroup_destroy(cgrp);
/*
* Remove @cgrp directory along with the base files. @cgrp has an
* extra ref on its kn.
--
1.9.1
^ permalink raw reply related [flat|nested] 13+ messages in thread
* [PATCH] cgroup: add tracepoints to track cgroup events
@ 2014-08-21 3:46 ` Andrea Righi
0 siblings, 0 replies; 13+ messages in thread
From: Andrea Righi @ 2014-08-21 3:46 UTC (permalink / raw)
To: Tejun Heo, Li Zefan, Steven Rostedt, Ingo Molnar
Cc: cgroups-u79uwXL29TY76Z2rM5mHXA,
linux-kernel-u79uwXL29TY76Z2rM5mHXA, Andrea Righi, Matt Heaton
This patch adds the following tracepoints:
o trace_cgroup_create when a new cgroup is created
o trace_cgroup_destroy when a cgroup is removed
o trace_cgroup_task_migrate when a task/thread is moved from a cgroup to another
The purpose of these tracepoints is to identify and help cgroup "managers" to
diagnose problems and detect when they are doing an excessive amount of work.
Signed-off-by: Matt Heaton <matt-oIIqvOZpAevzfdHfmsDf5w@public.gmane.org>
Signed-off-by: Andrea Righi <andrea-oIIqvOZpAevzfdHfmsDf5w@public.gmane.org>
---
include/trace/events/cgroup.h | 95 +++++++++++++++++++++++++++++++++++++++++++
kernel/cgroup.c | 14 ++++++-
2 files changed, 108 insertions(+), 1 deletion(-)
create mode 100644 include/trace/events/cgroup.h
diff --git a/include/trace/events/cgroup.h b/include/trace/events/cgroup.h
new file mode 100644
index 0000000..937b41e
--- /dev/null
+++ b/include/trace/events/cgroup.h
@@ -0,0 +1,95 @@
+#undef TRACE_SYSTEM
+#define TRACE_SYSTEM cgroup
+
+#if !defined(_TRACE_CGROUP_H) || defined(TRACE_HEADER_MULTI_READ)
+#define _TRACE_CGROUP_H
+
+#include <linux/cgroup.h>
+#include <linux/tracepoint.h>
+
+#define TRACE_CGROUP_PATH_MAX 256
+
+#ifdef CREATE_TRACE_POINTS
+static inline void cgroup_safe_path(struct cgroup *cgrp, char *buf,
+ size_t buflen)
+{
+ char *path = cgroup_path(cgrp, buf, buflen);
+ size_t len;
+
+ if (likely(path)) {
+ /* NOTE: path is always NULL terminated */
+ len = strlen(path);
+ memmove(buf, path, len);
+ buf[len] = '\0';
+ } else {
+ strncpy(buf, "(NULL)", buflen);
+ }
+}
+#endif
+
+TRACE_EVENT(cgroup_create,
+
+ TP_PROTO(struct cgroup *cgrp),
+
+ TP_ARGS(cgrp),
+
+ TP_STRUCT__entry(
+ __array(char, name, TRACE_CGROUP_PATH_MAX)
+ ),
+
+ TP_fast_assign(
+ cgroup_safe_path(cgrp, __entry->name, TRACE_CGROUP_PATH_MAX);
+ ),
+
+ TP_printk("%s", __entry->name)
+);
+
+TRACE_EVENT(cgroup_destroy,
+
+ TP_PROTO(struct cgroup *cgrp),
+
+ TP_ARGS(cgrp),
+
+ TP_STRUCT__entry(
+ __array(char, name, TRACE_CGROUP_PATH_MAX)
+ ),
+
+ TP_fast_assign(
+ cgroup_safe_path(cgrp, __entry->name, TRACE_CGROUP_PATH_MAX);
+ ),
+
+ TP_printk("%s", __entry->name)
+);
+
+TRACE_EVENT(cgroup_task_migrate,
+
+ TP_PROTO(struct cgroup *old_cgrp, struct cgroup *new_cgrp,
+ const struct task_struct *p),
+
+ TP_ARGS(old_cgrp, new_cgrp, p),
+
+ TP_STRUCT__entry(
+ __field(pid_t, pid)
+ __array(char, old_name, TRACE_CGROUP_PATH_MAX)
+ __array(char, new_name, TRACE_CGROUP_PATH_MAX)
+ __array(char, comm, TASK_COMM_LEN)
+ ),
+
+ TP_fast_assign(
+ __entry->pid = p->pid;
+ memcpy(__entry->comm, p->comm, TASK_COMM_LEN);
+ cgroup_safe_path(old_cgrp, __entry->old_name,
+ TRACE_CGROUP_PATH_MAX);
+ cgroup_safe_path(new_cgrp, __entry->new_name,
+ TRACE_CGROUP_PATH_MAX);
+ ),
+
+ TP_printk("pid=%d comm=%s from=%s to=%s",
+ __entry->pid, __entry->comm,
+ __entry->old_name, __entry->new_name)
+);
+
+#endif /* _TRACE_CGROUP_H */
+
+/* This part must be outside protection */
+#include <trace/define_trace.h>
diff --git a/kernel/cgroup.c b/kernel/cgroup.c
index 7dc8788..00a50b9 100644
--- a/kernel/cgroup.c
+++ b/kernel/cgroup.c
@@ -60,6 +60,9 @@
#include <linux/atomic.h>
+#define CREATE_TRACE_POINTS
+#include <trace/events/cgroup.h>
+
/*
* pidlists linger the following amount before being destroyed. The goal
* is avoiding frequent destruction in the middle of consecutive read calls
@@ -2014,6 +2017,7 @@ struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset)
* Must be called with cgroup_mutex, threadgroup and css_set_rwsem locked.
*/
static void cgroup_task_migrate(struct cgroup *old_cgrp,
+ struct cgroup *new_cgrp,
struct task_struct *tsk,
struct css_set *new_cset)
{
@@ -2022,6 +2026,8 @@ static void cgroup_task_migrate(struct cgroup *old_cgrp,
lockdep_assert_held(&cgroup_mutex);
lockdep_assert_held(&css_set_rwsem);
+ trace_cgroup_task_migrate(old_cgrp, new_cgrp, tsk);
+
/*
* We are synchronized through threadgroup_lock() against PF_EXITING
* setting such that we can't race against cgroup_exit() changing the
@@ -2274,7 +2280,7 @@ static int cgroup_migrate(struct cgroup *cgrp, struct task_struct *leader,
down_write(&css_set_rwsem);
list_for_each_entry(cset, &tset.src_csets, mg_node) {
list_for_each_entry_safe(task, tmp_task, &cset->mg_tasks, cg_list)
- cgroup_task_migrate(cset->mg_src_cgrp, task,
+ cgroup_task_migrate(cset->mg_src_cgrp, cgrp, task,
cset->mg_dst_cset);
}
up_write(&css_set_rwsem);
@@ -2988,6 +2994,7 @@ static int cgroup_rename(struct kernfs_node *kn, struct kernfs_node *new_parent,
if (cgroup_on_dfl(cgrp))
return -EPERM;
+ trace_cgroup_destroy(cgrp);
/*
* We're gonna grab cgroup_mutex which nests outside kernfs
* active_ref. kernfs_rename() doesn't require active_ref
@@ -3004,6 +3011,9 @@ static int cgroup_rename(struct kernfs_node *kn, struct kernfs_node *new_parent,
kernfs_unbreak_active_protection(kn);
kernfs_unbreak_active_protection(new_parent);
+
+ trace_cgroup_create(cgrp);
+
return ret;
}
@@ -4587,6 +4597,7 @@ static int cgroup_mkdir(struct kernfs_node *parent_kn, const char *name,
goto out_free_id;
}
cgrp->kn = kn;
+ trace_cgroup_create(cgrp);
/*
* This extra ref will be put in cgroup_free_fn() and guarantees
@@ -4791,6 +4802,7 @@ static int cgroup_destroy_locked(struct cgroup *cgrp)
list_del_init(&cgrp->release_list);
raw_spin_unlock(&release_list_lock);
+ trace_cgroup_destroy(cgrp);
/*
* Remove @cgrp directory along with the base files. @cgrp has an
* extra ref on its kn.
--
1.9.1
^ permalink raw reply related [flat|nested] 13+ messages in thread
* Re: [PATCH] cgroup: add tracepoints to track cgroup events
@ 2014-08-21 14:13 ` Tejun Heo
0 siblings, 0 replies; 13+ messages in thread
From: Tejun Heo @ 2014-08-21 14:13 UTC (permalink / raw)
To: Andrea Righi
Cc: Li Zefan, Steven Rostedt, Ingo Molnar, cgroups, linux-kernel,
Matt Heaton
Hello,
On Wed, Aug 20, 2014 at 09:46:25PM -0600, Andrea Righi wrote:
> This patch adds the following tracepoints:
> o trace_cgroup_create when a new cgroup is created
> o trace_cgroup_destroy when a cgroup is removed
> o trace_cgroup_task_migrate when a task/thread is moved from a cgroup to another
>
> The purpose of these tracepoints is to identify and help cgroup "managers" to
> diagnose problems and detect when they are doing an excessive amount of work.
Using TPs for this looks like a really roundabout way of doing this
when the whole interface is based on filesystem. Extending
kernfs_notity to support directory events seems like a better way to
do this.
Thanks.
--
tejun
^ permalink raw reply [flat|nested] 13+ messages in thread
* Re: [PATCH] cgroup: add tracepoints to track cgroup events
@ 2014-08-21 14:13 ` Tejun Heo
0 siblings, 0 replies; 13+ messages in thread
From: Tejun Heo @ 2014-08-21 14:13 UTC (permalink / raw)
To: Andrea Righi
Cc: Li Zefan, Steven Rostedt, Ingo Molnar,
cgroups-u79uwXL29TY76Z2rM5mHXA,
linux-kernel-u79uwXL29TY76Z2rM5mHXA, Matt Heaton
Hello,
On Wed, Aug 20, 2014 at 09:46:25PM -0600, Andrea Righi wrote:
> This patch adds the following tracepoints:
> o trace_cgroup_create when a new cgroup is created
> o trace_cgroup_destroy when a cgroup is removed
> o trace_cgroup_task_migrate when a task/thread is moved from a cgroup to another
>
> The purpose of these tracepoints is to identify and help cgroup "managers" to
> diagnose problems and detect when they are doing an excessive amount of work.
Using TPs for this looks like a really roundabout way of doing this
when the whole interface is based on filesystem. Extending
kernfs_notity to support directory events seems like a better way to
do this.
Thanks.
--
tejun
^ permalink raw reply [flat|nested] 13+ messages in thread
* Re: [PATCH] cgroup: add tracepoints to track cgroup events
@ 2014-08-21 15:35 ` Andrea Righi
0 siblings, 0 replies; 13+ messages in thread
From: Andrea Righi @ 2014-08-21 15:35 UTC (permalink / raw)
To: Tejun Heo
Cc: Li Zefan, Steven Rostedt, Ingo Molnar, cgroups, linux-kernel,
Matt Heaton
On Thu, Aug 21, 2014 at 09:13:30AM -0500, Tejun Heo wrote:
> Hello,
>
> On Wed, Aug 20, 2014 at 09:46:25PM -0600, Andrea Righi wrote:
> > This patch adds the following tracepoints:
> > o trace_cgroup_create when a new cgroup is created
> > o trace_cgroup_destroy when a cgroup is removed
> > o trace_cgroup_task_migrate when a task/thread is moved from a cgroup to another
> >
> > The purpose of these tracepoints is to identify and help cgroup "managers" to
> > diagnose problems and detect when they are doing an excessive amount of work.
>
> Using TPs for this looks like a really roundabout way of doing this
> when the whole interface is based on filesystem. Extending
> kernfs_notity to support directory events seems like a better way to
> do this.
>
> Thanks.
>
> --
> tejun
Agreed. Thanks for the suggestion, Tejun.
-Andrea
^ permalink raw reply [flat|nested] 13+ messages in thread
* Re: [PATCH] cgroup: add tracepoints to track cgroup events
@ 2014-08-21 15:35 ` Andrea Righi
0 siblings, 0 replies; 13+ messages in thread
From: Andrea Righi @ 2014-08-21 15:35 UTC (permalink / raw)
To: Tejun Heo
Cc: Li Zefan, Steven Rostedt, Ingo Molnar,
cgroups-u79uwXL29TY76Z2rM5mHXA,
linux-kernel-u79uwXL29TY76Z2rM5mHXA, Matt Heaton
On Thu, Aug 21, 2014 at 09:13:30AM -0500, Tejun Heo wrote:
> Hello,
>
> On Wed, Aug 20, 2014 at 09:46:25PM -0600, Andrea Righi wrote:
> > This patch adds the following tracepoints:
> > o trace_cgroup_create when a new cgroup is created
> > o trace_cgroup_destroy when a cgroup is removed
> > o trace_cgroup_task_migrate when a task/thread is moved from a cgroup to another
> >
> > The purpose of these tracepoints is to identify and help cgroup "managers" to
> > diagnose problems and detect when they are doing an excessive amount of work.
>
> Using TPs for this looks like a really roundabout way of doing this
> when the whole interface is based on filesystem. Extending
> kernfs_notity to support directory events seems like a better way to
> do this.
>
> Thanks.
>
> --
> tejun
Agreed. Thanks for the suggestion, Tejun.
-Andrea
^ permalink raw reply [flat|nested] 13+ messages in thread
* Re: [PATCH] cgroup: add tracepoints to track cgroup events
2014-08-21 15:35 ` Andrea Righi
(?)
@ 2014-08-21 17:00 ` Andrea Righi
2014-08-21 17:07 ` Tejun Heo
-1 siblings, 1 reply; 13+ messages in thread
From: Andrea Righi @ 2014-08-21 17:00 UTC (permalink / raw)
To: Tejun Heo
Cc: Li Zefan, Steven Rostedt, Ingo Molnar, cgroups, linux-kernel,
Matt Heaton
On Thu, Aug 21, 2014 at 09:35:41AM -0600, Andrea Righi wrote:
> On Thu, Aug 21, 2014 at 09:13:30AM -0500, Tejun Heo wrote:
> > Hello,
> >
> > On Wed, Aug 20, 2014 at 09:46:25PM -0600, Andrea Righi wrote:
> > > This patch adds the following tracepoints:
> > > o trace_cgroup_create when a new cgroup is created
> > > o trace_cgroup_destroy when a cgroup is removed
> > > o trace_cgroup_task_migrate when a task/thread is moved from a cgroup to another
> > >
> > > The purpose of these tracepoints is to identify and help cgroup "managers" to
> > > diagnose problems and detect when they are doing an excessive amount of work.
> >
> > Using TPs for this looks like a really roundabout way of doing this
> > when the whole interface is based on filesystem. Extending
> > kernfs_notity to support directory events seems like a better way to
> > do this.
> >
> > Thanks.
> >
> > --
> > tejun
>
> Agreed. Thanks for the suggestion, Tejun.
>
> -Andrea
hmm... am I missing something or we already support directory events?
Example:
root@Dell:~# grep cgroups /proc/mounts
none /cgroups cgroup rw,relatime,cpuset,cpu,cpuacct,memory,devices,freezer,perf_event,hugetlb 0 0
root@Dell:~# inotifywait -m -r -e modify -e move -e create -e delete /cgroups
Setting up watches. Beware: since -r was given, this may take a while!
Watches established.
/cgroups/ CREATE,ISDIR test
/cgroups/test/ MODIFY cgroup.procs
/cgroups/test/ MODIFY cgroup.procs
/cgroups/test/ MODIFY cgroup.populated
/cgroups/ MODIFY cgroup.procs
/cgroups/ MODIFY cgroup.procs
/cgroups/test/ MODIFY cgroup.populated
/cgroups/ DELETE,ISDIR test
I still need to figure out a smart way to track which PIDs are
added/removed to/from cgroup.procs from userland (inotifywait + git? :)),
but all the other informations provided by my tracepoint patch seem to
be already available via [di]notify.
Thanks,
-Andrea
^ permalink raw reply [flat|nested] 13+ messages in thread
* Re: [PATCH] cgroup: add tracepoints to track cgroup events
2014-08-21 17:00 ` Andrea Righi
@ 2014-08-21 17:07 ` Tejun Heo
2014-08-21 17:45 ` Andrea Righi
2014-08-21 23:45 ` Steven Rostedt
0 siblings, 2 replies; 13+ messages in thread
From: Tejun Heo @ 2014-08-21 17:07 UTC (permalink / raw)
To: Andrea Righi
Cc: Li Zefan, Steven Rostedt, Ingo Molnar, cgroups, linux-kernel,
Matt Heaton
Hello, Anrea.
On Thu, Aug 21, 2014 at 11:00:02AM -0600, Andrea Righi wrote:
> hmm... am I missing something or we already support directory events?
Ah, right, those mkdir/rmdir and writes automatically generate those
events.
> root@Dell:~# grep cgroups /proc/mounts
> none /cgroups cgroup rw,relatime,cpuset,cpu,cpuacct,memory,devices,freezer,perf_event,hugetlb 0 0
> root@Dell:~# inotifywait -m -r -e modify -e move -e create -e delete /cgroups
> Setting up watches. Beware: since -r was given, this may take a while!
> Watches established.
> /cgroups/ CREATE,ISDIR test
> /cgroups/test/ MODIFY cgroup.procs
> /cgroups/test/ MODIFY cgroup.procs
> /cgroups/test/ MODIFY cgroup.populated
> /cgroups/ MODIFY cgroup.procs
> /cgroups/ MODIFY cgroup.procs
> /cgroups/test/ MODIFY cgroup.populated
> /cgroups/ DELETE,ISDIR test
>
> I still need to figure out a smart way to track which PIDs are
> added/removed to/from cgroup.procs from userland (inotifywait + git? :)),
> but all the other informations provided by my tracepoint patch seem to
> be already available via [di]notify.
Hmmm... yeah, determining exactly which pids got added / removed can
be cumbersome from just MODIFY events. That said, what are you trying
to do with such information?
Thanks.
--
tejun
^ permalink raw reply [flat|nested] 13+ messages in thread
* Re: [PATCH] cgroup: add tracepoints to track cgroup events
@ 2014-08-21 17:45 ` Andrea Righi
0 siblings, 0 replies; 13+ messages in thread
From: Andrea Righi @ 2014-08-21 17:45 UTC (permalink / raw)
To: Tejun Heo
Cc: Li Zefan, Steven Rostedt, Ingo Molnar, cgroups, linux-kernel,
Matt Heaton
On Thu, Aug 21, 2014 at 12:07:01PM -0500, Tejun Heo wrote:
...
> > I still need to figure out a smart way to track which PIDs are
> > added/removed to/from cgroup.procs from userland (inotifywait + git? :)),
> > but all the other informations provided by my tracepoint patch seem to
> > be already available via [di]notify.
>
> Hmmm... yeah, determining exactly which pids got added / removed can
> be cumbersome from just MODIFY events. That said, what are you trying
> to do with such information?
Just trying to figure out if a particular application/pid is moved too
much across different cgroups, or collect per-pid / per-cgroup
statistics in general.
Thanks,
-Andrea
^ permalink raw reply [flat|nested] 13+ messages in thread
* Re: [PATCH] cgroup: add tracepoints to track cgroup events
@ 2014-08-21 17:45 ` Andrea Righi
0 siblings, 0 replies; 13+ messages in thread
From: Andrea Righi @ 2014-08-21 17:45 UTC (permalink / raw)
To: Tejun Heo
Cc: Li Zefan, Steven Rostedt, Ingo Molnar,
cgroups-u79uwXL29TY76Z2rM5mHXA,
linux-kernel-u79uwXL29TY76Z2rM5mHXA, Matt Heaton
On Thu, Aug 21, 2014 at 12:07:01PM -0500, Tejun Heo wrote:
...
> > I still need to figure out a smart way to track which PIDs are
> > added/removed to/from cgroup.procs from userland (inotifywait + git? :)),
> > but all the other informations provided by my tracepoint patch seem to
> > be already available via [di]notify.
>
> Hmmm... yeah, determining exactly which pids got added / removed can
> be cumbersome from just MODIFY events. That said, what are you trying
> to do with such information?
Just trying to figure out if a particular application/pid is moved too
much across different cgroups, or collect per-pid / per-cgroup
statistics in general.
Thanks,
-Andrea
^ permalink raw reply [flat|nested] 13+ messages in thread
* Re: [PATCH] cgroup: add tracepoints to track cgroup events
2014-08-21 17:07 ` Tejun Heo
2014-08-21 17:45 ` Andrea Righi
@ 2014-08-21 23:45 ` Steven Rostedt
2014-08-22 4:05 ` Andrea Righi
1 sibling, 1 reply; 13+ messages in thread
From: Steven Rostedt @ 2014-08-21 23:45 UTC (permalink / raw)
To: Tejun Heo
Cc: Andrea Righi, Li Zefan, Ingo Molnar, cgroups, linux-kernel, Matt Heaton
On Thu, 21 Aug 2014 12:07:01 -0500
Tejun Heo <tj@kernel.org> wrote:
> Hello, Anrea.
>
> On Thu, Aug 21, 2014 at 11:00:02AM -0600, Andrea Righi wrote:
> > hmm... am I missing something or we already support directory events?
>
> Ah, right, those mkdir/rmdir and writes automatically generate those
> events.
>
> > root@Dell:~# grep cgroups /proc/mounts
> > none /cgroups cgroup rw,relatime,cpuset,cpu,cpuacct,memory,devices,freezer,perf_event,hugetlb 0 0
> > root@Dell:~# inotifywait -m -r -e modify -e move -e create -e delete /cgroups
> > Setting up watches. Beware: since -r was given, this may take a while!
> > Watches established.
> > /cgroups/ CREATE,ISDIR test
> > /cgroups/test/ MODIFY cgroup.procs
> > /cgroups/test/ MODIFY cgroup.procs
> > /cgroups/test/ MODIFY cgroup.populated
> > /cgroups/ MODIFY cgroup.procs
> > /cgroups/ MODIFY cgroup.procs
> > /cgroups/test/ MODIFY cgroup.populated
> > /cgroups/ DELETE,ISDIR test
> >
> > I still need to figure out a smart way to track which PIDs are
> > added/removed to/from cgroup.procs from userland (inotifywait + git? :)),
> > but all the other informations provided by my tracepoint patch seem to
> > be already available via [di]notify.
>
> Hmmm... yeah, determining exactly which pids got added / removed can
> be cumbersome from just MODIFY events. That said, what are you trying
> to do with such information?
>
OK, is this patch not being pushed then? I have a lot of comments to
make about it, but if this patch is being dropped for another way of
doing things I wont waste my time on it.
Thanks,
-- Steve
^ permalink raw reply [flat|nested] 13+ messages in thread
* Re: [PATCH] cgroup: add tracepoints to track cgroup events
@ 2014-08-22 4:05 ` Andrea Righi
0 siblings, 0 replies; 13+ messages in thread
From: Andrea Righi @ 2014-08-22 4:05 UTC (permalink / raw)
To: Steven Rostedt
Cc: Tejun Heo, Li Zefan, Ingo Molnar, cgroups, linux-kernel, Matt Heaton
On Thu, Aug 21, 2014 at 07:45:14PM -0400, Steven Rostedt wrote:
> On Thu, 21 Aug 2014 12:07:01 -0500
> Tejun Heo <tj@kernel.org> wrote:
>
> > Hello, Anrea.
> >
> > On Thu, Aug 21, 2014 at 11:00:02AM -0600, Andrea Righi wrote:
> > > hmm... am I missing something or we already support directory events?
> >
> > Ah, right, those mkdir/rmdir and writes automatically generate those
> > events.
> >
> > > root@Dell:~# grep cgroups /proc/mounts
> > > none /cgroups cgroup rw,relatime,cpuset,cpu,cpuacct,memory,devices,freezer,perf_event,hugetlb 0 0
> > > root@Dell:~# inotifywait -m -r -e modify -e move -e create -e delete /cgroups
> > > Setting up watches. Beware: since -r was given, this may take a while!
> > > Watches established.
> > > /cgroups/ CREATE,ISDIR test
> > > /cgroups/test/ MODIFY cgroup.procs
> > > /cgroups/test/ MODIFY cgroup.procs
> > > /cgroups/test/ MODIFY cgroup.populated
> > > /cgroups/ MODIFY cgroup.procs
> > > /cgroups/ MODIFY cgroup.procs
> > > /cgroups/test/ MODIFY cgroup.populated
> > > /cgroups/ DELETE,ISDIR test
> > >
> > > I still need to figure out a smart way to track which PIDs are
> > > added/removed to/from cgroup.procs from userland (inotifywait + git? :)),
> > > but all the other informations provided by my tracepoint patch seem to
> > > be already available via [di]notify.
> >
> > Hmmm... yeah, determining exactly which pids got added / removed can
> > be cumbersome from just MODIFY events. That said, what are you trying
> > to do with such information?
> >
>
> OK, is this patch not being pushed then? I have a lot of comments to
> make about it, but if this patch is being dropped for another way of
> doing things I wont waste my time on it.
>
> Thanks,
>
> -- Steve
Comments are always welcome, but at this point I'd say we can drop this
patch, so don't waste your time on it. I can find an alternative way to
get the same informations from user-space.
Thanks,
-Andrea
^ permalink raw reply [flat|nested] 13+ messages in thread
* Re: [PATCH] cgroup: add tracepoints to track cgroup events
@ 2014-08-22 4:05 ` Andrea Righi
0 siblings, 0 replies; 13+ messages in thread
From: Andrea Righi @ 2014-08-22 4:05 UTC (permalink / raw)
To: Steven Rostedt
Cc: Tejun Heo, Li Zefan, Ingo Molnar, cgroups-u79uwXL29TY76Z2rM5mHXA,
linux-kernel-u79uwXL29TY76Z2rM5mHXA, Matt Heaton
On Thu, Aug 21, 2014 at 07:45:14PM -0400, Steven Rostedt wrote:
> On Thu, 21 Aug 2014 12:07:01 -0500
> Tejun Heo <tj-DgEjT+Ai2ygdnm+yROfE0A@public.gmane.org> wrote:
>
> > Hello, Anrea.
> >
> > On Thu, Aug 21, 2014 at 11:00:02AM -0600, Andrea Righi wrote:
> > > hmm... am I missing something or we already support directory events?
> >
> > Ah, right, those mkdir/rmdir and writes automatically generate those
> > events.
> >
> > > root@Dell:~# grep cgroups /proc/mounts
> > > none /cgroups cgroup rw,relatime,cpuset,cpu,cpuacct,memory,devices,freezer,perf_event,hugetlb 0 0
> > > root@Dell:~# inotifywait -m -r -e modify -e move -e create -e delete /cgroups
> > > Setting up watches. Beware: since -r was given, this may take a while!
> > > Watches established.
> > > /cgroups/ CREATE,ISDIR test
> > > /cgroups/test/ MODIFY cgroup.procs
> > > /cgroups/test/ MODIFY cgroup.procs
> > > /cgroups/test/ MODIFY cgroup.populated
> > > /cgroups/ MODIFY cgroup.procs
> > > /cgroups/ MODIFY cgroup.procs
> > > /cgroups/test/ MODIFY cgroup.populated
> > > /cgroups/ DELETE,ISDIR test
> > >
> > > I still need to figure out a smart way to track which PIDs are
> > > added/removed to/from cgroup.procs from userland (inotifywait + git? :)),
> > > but all the other informations provided by my tracepoint patch seem to
> > > be already available via [di]notify.
> >
> > Hmmm... yeah, determining exactly which pids got added / removed can
> > be cumbersome from just MODIFY events. That said, what are you trying
> > to do with such information?
> >
>
> OK, is this patch not being pushed then? I have a lot of comments to
> make about it, but if this patch is being dropped for another way of
> doing things I wont waste my time on it.
>
> Thanks,
>
> -- Steve
Comments are always welcome, but at this point I'd say we can drop this
patch, so don't waste your time on it. I can find an alternative way to
get the same informations from user-space.
Thanks,
-Andrea
^ permalink raw reply [flat|nested] 13+ messages in thread
end of thread, other threads:[~2014-08-22 4:05 UTC | newest]
Thread overview: 13+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2014-08-21 3:46 [PATCH] cgroup: add tracepoints to track cgroup events Andrea Righi
2014-08-21 3:46 ` Andrea Righi
2014-08-21 14:13 ` Tejun Heo
2014-08-21 14:13 ` Tejun Heo
2014-08-21 15:35 ` Andrea Righi
2014-08-21 15:35 ` Andrea Righi
2014-08-21 17:00 ` Andrea Righi
2014-08-21 17:07 ` Tejun Heo
2014-08-21 17:45 ` Andrea Righi
2014-08-21 17:45 ` Andrea Righi
2014-08-21 23:45 ` Steven Rostedt
2014-08-22 4:05 ` Andrea Righi
2014-08-22 4:05 ` Andrea Righi
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.