Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Date:   Mon, 31 Jan 2022 16:26:21 +0100
From:   "Dr. Thomas Orgis" <thomas.orgis@uni-hamburg.de>
To:     Balbir Singh <bsingharora@gmail.com>
CC:     <linux-kernel@vger.kernel.org>
Subject: Re: **SPAM** [PATCH 1/2] taskstats: version 11 with tgid
Message-ID: <20220131162621.11fc8f59@plasteblaster>
In-Reply-To: <20220110194417.147d4eff@plasteblaster>
References: <20220110194417.147d4eff@plasteblaster>
Organization: =?UTF-8?B?VW5pdmVyc2l0w6R0?= Hamburg
MIME-Version: 1.0
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: 8BIT
Precedence: bulk

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" <thomas.orgis@uni-hamburg.de>: 

> From 9d3d915c4e0c1e4ff3a54d73851cedb613c7df44 Mon Sep 17 00:00:00 2001
> From: "Dr. Thomas Orgis" <thomas.orgis@uni-hamburg.de>
> 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 <thomas.orgis@uni-hamburg.de>
> ---
>  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