From 9d3d915c4e0c1e4ff3a54d73851cedb613c7df44 Mon Sep 17 00:00:00 2001
From: "Dr. Thomas Orgis" <[email protected]>
Date: Mon, 10 Jan 2022 17:10:41 +0100
Subject: [PATCH 1/2] taskstats: version 11 with tgid
This adds the ac_tgid field to the taskstats struct, to be able to tell
apart (additional) threads and processes when a client application monitors
task exit events.
I need this for giving HPC users an overview over how many instances of
which programs they ran and how much resources each one used, including
the distinction between multithreaded programs (parallelized scientific
applications) and many process instances (script calling lots of
short-running programs). It is nice to get this information accurately via
taskstats with a data set once for each task that exits, as opposed to
expensive/inaccurate sampling of any sort.
There are process ID and parent process ID already in the struct. The
thread group ID seems like something that should accompany those, anyway.
base-commit: df0cc57e057f18e44dac8e6c18aba47ab53202f9
Signed-off-by: Dr. Thomas Orgis <[email protected]>
---
include/uapi/linux/taskstats.h | 4 +++-
kernel/tsacct.c | 1 +
2 files changed, 4 insertions(+), 1 deletion(-)
diff --git a/include/uapi/linux/taskstats.h b/include/uapi/linux/taskstats.h
index ccbd08709321..9248dc6bcb4a 100644
--- a/include/uapi/linux/taskstats.h
+++ b/include/uapi/linux/taskstats.h
@@ -34,7 +34,7 @@
*/
-#define TASKSTATS_VERSION 10
+#define TASKSTATS_VERSION 11
#define TS_COMM_LEN 32 /* should be >= TASK_COMM_LEN
* in linux/sched.h */
@@ -172,6 +172,8 @@ struct taskstats {
/* v10: 64-bit btime to avoid overflow */
__u64 ac_btime64; /* 64-bit begin time */
+ /* v11: thread group ID to identify process vs. (non-leader) thread */
+ __u32 ac_tgid;
};
diff --git a/kernel/tsacct.c b/kernel/tsacct.c
index f00de83d0246..959ae3a26f1b 100644
--- a/kernel/tsacct.c
+++ b/kernel/tsacct.c
@@ -52,6 +52,7 @@ void bacct_add_tsk(struct user_namespace *user_ns,
stats->ac_nice = task_nice(tsk);
stats->ac_sched = tsk->policy;
stats->ac_pid = task_pid_nr_ns(tsk, pid_ns);
+ stats->ac_tgid = task_tgid_nr_ns(tsk, pid_ns);
rcu_read_lock();
tcred = __task_cred(tsk);
stats->ac_uid = from_kuid_munged(user_ns, tcred->uid);
base-commit: df0cc57e057f18e44dac8e6c18aba47ab53202f9
--
2.29.2
--
Dr. Thomas Orgis
HPC @ Universität Hamburg
Dear Andrew Morton,
I am trying to get my taskstats updates noticed since 2018 now. Balbir
Singh does not seem to be active/responding at the mail address that
scripts/get_maintainer.pl kernel/taskstats.c
produces. The guide at
https://docs.kernel.org/process/submitting-patches.html
suggests that I should bother you directly in that case. May I kindly
point to the two patches that I'd like to have reviewed and hopefully
included?
These are the two patches in the LKML archive:
https://lkml.org/lkml/2022/1/10/1136
https://lkml.org/lkml/2022/1/10/1135
I hope they are trivial enough, in the end.
Alrighty then,
Thomas
--
Dr. Thomas Orgis
HPC @ Universität Hamburg
Hi,
any hints on this one? Should I group this into one patch with the
other one to increase the struct version only once? I figured that the
other changes in patch 2 are more controversial and need more discussion.
Or should I rather work on a taskstats mode that consinstently returns
TGID stats (and possibly only those) with accumulated fields like CPU
time for the whole process? This is more like what I am after in the
end (plus a task count for this process to tell if it was
multithreaded), but just providing the tgid enables userspace to do
that accumulation.
Or should I abandon taskstats in total and look elsewhere? So far I did
not see a place that gives me equivalent data (e.g. the limited fields
of BSD process accounting).
Am Mon, 10 Jan 2022 19:44:17 +0100
schrieb "Dr. Thomas Orgis" <[email protected]>:
> From 9d3d915c4e0c1e4ff3a54d73851cedb613c7df44 Mon Sep 17 00:00:00 2001
> From: "Dr. Thomas Orgis" <[email protected]>
> Date: Mon, 10 Jan 2022 17:10:41 +0100
> Subject: [PATCH 1/2] taskstats: version 11 with tgid
>
> This adds the ac_tgid field to the taskstats struct, to be able to tell
> apart (additional) threads and processes when a client application monitors
> task exit events.
>
> I need this for giving HPC users an overview over how many instances of
> which programs they ran and how much resources each one used, including
> the distinction between multithreaded programs (parallelized scientific
> applications) and many process instances (script calling lots of
> short-running programs). It is nice to get this information accurately via
> taskstats with a data set once for each task that exits, as opposed to
> expensive/inaccurate sampling of any sort.
>
> There are process ID and parent process ID already in the struct. The
> thread group ID seems like something that should accompany those, anyway.
>
> base-commit: df0cc57e057f18e44dac8e6c18aba47ab53202f9
>
> Signed-off-by: Dr. Thomas Orgis <[email protected]>
> ---
> include/uapi/linux/taskstats.h | 4 +++-
> kernel/tsacct.c | 1 +
> 2 files changed, 4 insertions(+), 1 deletion(-)
>
> diff --git a/include/uapi/linux/taskstats.h b/include/uapi/linux/taskstats.h
> index ccbd08709321..9248dc6bcb4a 100644
> --- a/include/uapi/linux/taskstats.h
> +++ b/include/uapi/linux/taskstats.h
> @@ -34,7 +34,7 @@
> */
>
>
> -#define TASKSTATS_VERSION 10
> +#define TASKSTATS_VERSION 11
> #define TS_COMM_LEN 32 /* should be >= TASK_COMM_LEN
> * in linux/sched.h */
>
> @@ -172,6 +172,8 @@ struct taskstats {
>
> /* v10: 64-bit btime to avoid overflow */
> __u64 ac_btime64; /* 64-bit begin time */
> + /* v11: thread group ID to identify process vs. (non-leader) thread */
> + __u32 ac_tgid;
> };
>
>
> diff --git a/kernel/tsacct.c b/kernel/tsacct.c
> index f00de83d0246..959ae3a26f1b 100644
> --- a/kernel/tsacct.c
> +++ b/kernel/tsacct.c
> @@ -52,6 +52,7 @@ void bacct_add_tsk(struct user_namespace *user_ns,
> stats->ac_nice = task_nice(tsk);
> stats->ac_sched = tsk->policy;
> stats->ac_pid = task_pid_nr_ns(tsk, pid_ns);
> + stats->ac_tgid = task_tgid_nr_ns(tsk, pid_ns);
> rcu_read_lock();
> tcred = __task_cred(tsk);
> stats->ac_uid = from_kuid_munged(user_ns, tcred->uid);
>
> base-commit: df0cc57e057f18e44dac8e6c18aba47ab53202f9
> --
> 2.29.2
Alrighty then,
Thomas
--
Dr. Thomas Orgis
HPC @ Universität Hamburg