Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
Date:   Wed, 26 Feb 2020 17:19:41 +0100
From:   Peter Zijlstra <peterz@infradead.org>
To:     Luigi Rizzo <lrizzo@google.com>
Cc:     linux-kernel@vger.kernel.org, mhiramat@kernel.org,
        akpm@linux-foundation.org, gregkh@linuxfoundation.org,
        naveen.n.rao@linux.ibm.com, ardb@kernel.org, rizzo@iet.unipi.it,
        pabeni@redhat.com, giuseppe.lettieri@unipi.it, toke@redhat.com,
        hawk@kernel.org, mingo@redhat.com, acme@kernel.org,
        rostedt@goodmis.org
Subject: Re: [PATCH v2 1/2] kstats: kernel metric collector
Message-ID: <20200226161941.GZ18400@hirez.programming.kicks-ass.net>
References: <20200226134637.31670-1-lrizzo@google.com>
 <20200226134637.31670-2-lrizzo@google.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20200226134637.31670-2-lrizzo@google.com>
User-Agent: Mutt/1.10.1 (2018-07-13)
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Wed, Feb 26, 2020 at 05:46:36AM -0800, Luigi Rizzo wrote:
> kstats is a helper to accumulate in-kernel metrics (timestamps, sizes,
> etc.) and show distributions through debugfs.
> Set CONFIG_KSTATS=m or y to enable it.
> 
> Creating a metric takes one line of code (and one to destroy it):
> 
>   struct kstats *key = kstats_new("foo", 3 /* frac_bits */);
>   ...
>   kstats_delete(key);
> 
> The following line records a u64 sample:
> 
>   kstats_record(key, value);
> 
> kstats_record() is cheap (5ns hot cache, 250ns cold cache). Samples are
> accumulated in a per-cpu array with 2^frac_bits slots for each power
> of 2. Using frac_bits = 3 gives about 30 slots per decade.

So I think everybody + dog has written code like this, although I never
bothered with the log2 based buckets myself. Nor have I ever bothered
with doing a debugfs interface.

I find it very hard to convince myself something like this deserves to
live upstream, vs. remaining in the local debug/hack toolbox.

Tracing has an aggregator (histogram), you can dump the raw deltas, or
you can hack up a custom aggregator in a few lines, or you do BPF if
you're so inclined.

Why do we need this specific one?