From owner-freebsd-virtualization@freebsd.org Tue Aug 28 13:37:21 2018 Return-Path: Delivered-To: freebsd-virtualization@mailman.ysv.freebsd.org Received: from mx1.freebsd.org (mx1.freebsd.org [IPv6:2610:1c1:1:606c::19:1]) by mailman.ysv.freebsd.org (Postfix) with ESMTP id 8C15C108C936 for ; Tue, 28 Aug 2018 13:37:21 +0000 (UTC) (envelope-from freebsd-rwg@pdx.rh.CN85.dnsmgr.net) Received: from pdx.rh.CN85.dnsmgr.net (br1.CN84in.dnsmgr.net [69.59.192.140]) (using TLSv1 with cipher DHE-RSA-AES256-SHA (256/256 bits)) (Client did not present a certificate) by mx1.freebsd.org (Postfix) with ESMTPS id 0815A842B7 for ; Tue, 28 Aug 2018 13:37:20 +0000 (UTC) (envelope-from freebsd-rwg@pdx.rh.CN85.dnsmgr.net) Received: from pdx.rh.CN85.dnsmgr.net (localhost [127.0.0.1]) by pdx.rh.CN85.dnsmgr.net (8.13.3/8.13.3) with ESMTP id w7SDbB3G015574; Tue, 28 Aug 2018 06:37:11 -0700 (PDT) (envelope-from freebsd-rwg@pdx.rh.CN85.dnsmgr.net) Received: (from freebsd-rwg@localhost) by pdx.rh.CN85.dnsmgr.net (8.13.3/8.13.3/Submit) id w7SDbAEe015573; Tue, 28 Aug 2018 06:37:11 -0700 (PDT) (envelope-from freebsd-rwg) From: "Rodney W. Grimes" Message-Id: <201808281337.w7SDbAEe015573@pdx.rh.CN85.dnsmgr.net> Subject: Re: on bhyve statistics In-Reply-To: To: Anish Date: Tue, 28 Aug 2018 06:37:10 -0700 (PDT) CC: Fabian Freyer , "freebsd-virtualization@freebsd.org" X-Mailer: ELM [version 2.4ME+ PL121h (25)] MIME-Version: 1.0 Content-Transfer-Encoding: 7bit Content-Type: text/plain; charset=US-ASCII X-BeenThere: freebsd-virtualization@freebsd.org X-Mailman-Version: 2.1.27 Precedence: list List-Id: "Discussion of various virtualization techniques FreeBSD supports." List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-List-Received-Date: Tue, 28 Aug 2018 13:37:21 -0000 > >> Currently, bhyve does not expose any of these statistics. All the stats > > available through bhyvectl --get-stats seem to be coming from the VMM, > > not from the userspace emulation. > > >That is correct, byhvectl is a diagnostics tool for getting > information from the kernel/vmm module. > > bhyvectl provide stats related to processor vmx/svm from vmm.ko and is the > first thing you want to run for performance regression. It will be nice to > include it as part of bhyve perf tool/dashboard that you are intended to > build. >From conversations with Peter Grehan he expressed that bhyvectl is purely a diagnostics tool that should not be depended on by any other tools. If you want to do similiar things you should program to the libvmmapi interface, not bhyvectl. > -Anish > > On Mon, Aug 27, 2018 at 8:20 AM Rodney W. Grimes < > freebsd-rwg@pdx.rh.cn85.dnsmgr.net> wrote: > > > > Hi list, > > > > > > I'm currently looking at getting the libvirt prometheus exporter[1] to > > > work with libvirt+bhyve. In its current state this doesn't work because > > > at least one of the API calls exposed by libvirt isn't implemented by > > > the libvirt bhyve driver - so I started looking at implementing it. > > > > > > The first API call in question is virDomainBlockStats[2], which returns > > > statistics (number of read and written bytes and ops, respectively). > > > > > > Currently, bhyve does not expose any of these statistics. All the stats > > > available through bhyvectl --get-stats seem to be coming from the VMM, > > > not from the userspace emulation. > > > > That is correct, byhvectl is a diagnostics tool for getting > > information from the kernel/vmm module. > > > > > OTOH, I did see that there are *some* > > > stats being collected in bhyverun.c (see struct bhyvestats {...} > > > stats;). I can't see how these are exposed though - a grep of /usr/src > > > turned up no other uses. Which brings me to the following questions: > > > > > > - are the stats in struct bhyvestats {...} stats exposed or used in any > > > non-obvious way? > > > > Not that I am aware of. > > > > > - architecturally, what would be the best ways to get stats out of the > > > user-space emulations? Off of the top of my head, I could think of the > > > following possibilities: > > > - prometheus exporter > > > - having some socket or pipe to request them > > > - DTrace probes > > > > > > I wouldn't mind implementing any of the above, and so would like to know > > > which of these (or other options) would be the most acceptable, and > > > would appreciate some guidance. > > > > I differ to others on what may be the best way to do this. > > > > > CC'ing novel@ for the libvirt side, and grehan@ for the architectural > > > bhyve questions. > > > > You should replace @grehan with @jhb,@tychon as Peter has moved on, > > and John and Tycho are now the bhyve maintainers. I was going to > > add them, and remove Peter, but I see no cc: anyway, so I am sure > > that they are on the virtualization list though. > > > > > Fabian > > > > > > [1] https://github.com/kumina/libvirt_exporter > > > [2] > > https://libvirt.org/html/libvirt-libvirt-domain.html#virDomainBlockStats > > -- > > Rod Grimes > > rgrimes@freebsd.org -- Rod Grimes rgrimes@freebsd.org