Re: [PATCH v9 0/5] KVM statistics data fd-based binary interface

From: Paolo Bonzini <pbonzini@redhat.com>
To: Leon Romanovsky <leon@kernel.org>
Cc: Jing Zhang <jingzhangos@google.com>, KVM <kvm@vger.kernel.org>,
	KVMARM <kvmarm@lists.cs.columbia.edu>,
	LinuxMIPS <linux-mips@vger.kernel.org>,
	KVMPPC <kvm-ppc@vger.kernel.org>,
	LinuxS390 <linux-s390@vger.kernel.org>,
	Linuxkselftest <linux-kselftest@vger.kernel.org>,
	Marc Zyngier <maz@kernel.org>, James Morse <james.morse@arm.com>,
	Julien Thierry <julien.thierry.kdev@gmail.com>,
	Suzuki K Poulose <suzuki.poulose@arm.com>,
	Will Deacon <will@kernel.org>,
	Huacai Chen <chenhuacai@kernel.org>,
	Aleksandar Markovic <aleksandar.qemu.devel@gmail.com>,
	Thomas Bogendoerfer <tsbogend@alpha.franken.de>,
	Paul Mackerras <paulus@ozlabs.org>,
	Christian Borntraeger <borntraeger@de.ibm.com>,
	Janosch Frank <frankja@linux.ibm.com>,
	David Hildenbrand <david@redhat.com>,
	Cornelia Huck <cohuck@redhat.com>,
	Claudio Imbrenda <imbrenda@linux.ibm.com>,
	Sean Christopherson <seanjc@google.com>,
	Vitaly Kuznetsov <vkuznets@redhat.com>,
	Jim Mattson <jmattson@google.com>,
	Peter Shier <pshier@google.com>, Oliver Upton <oupton@google.com>,
	David Rientjes <rientjes@google.com>,
	Emanuele Giuseppe Esposito <eesposit@redhat.com>,
	David Matlack <dmatlack@google.com>,
	Ricardo Koller <ricarkol@google.com>,
	Krish Sadhukhan <krish.sadhukhan@oracle.com>,
	Fuad Tabba <tabba@google.com>
Subject: Re: [PATCH v9 0/5] KVM statistics data fd-based binary interface
Date: Tue, 15 Jun 2021 13:03:34 +0200	[thread overview]
Message-ID: <9df462c0-e0ea-8173-0705-369d6a81107c@redhat.com> (raw)
In-Reply-To: <YMhcek2cIu3Oz5Ek@unreal>

On 15/06/21 09:53, Leon Romanovsky wrote:
>> Sorry for my naive questions, but how does telemetry get statistics
>> for hypervisors? Why is KVM different from hypervisors or NIC's statistics
>> or any other high speed devices (RDMA) that generate tons of data?
>
> So the answer to the question "why KVM is different" is that it doesn't
> have any stable identification except file descriptor. While hypervisors
> have stable names, NICs and RDMA devices have interface indexes etc.
> Did I get it right?

Right.

> And this was second part of my question, the first part was my attempt to
> get on answer why current statistics like process info (/proc/xxx/*), NICs
> (netlink) and RDMA (sysfs) are not using binary format.

NICs are using binary format (partly in struct ethtool_stats, partly in 
an array of u64).  For KVM we decided to put the schema and the stats in 
the same file (though you can use pread to get only the stats) to have a 
single interface and avoid ioctls, unlike having both ETH_GSTRINGS and 
ETH_GSTATS.

I wouldn't say processes are using any specific format.  There's a mix 
of "one value per file" (e.g. cpuset), human-readable tabular format 
(e.g. limits, sched), human- and machine-readable tabular format (e.g. 
status), and files that are ASCII but not human-readable (e.g. stat).

Paolo