Some oxenstored improvements (v3)
This is mainly about removing the 1024 fds limitation in the current oxenstored implementation. We also fixed some bugs and made some perf improvements along the way.
This is v3. The first 6 patches are the same as in v2. For patch 7/8/9, we
* Add a safe net mechanism for ill-behaved legacy clients
* Make some performance improvement to reduce syslog workload
* Small refactoring on patch 7/8 to accomodate the new changes
Cheers,
Zheng
PATCH v3 1/9 oxenstored: add a poll-based select mechanism
Currently, oxenstored uses Unix.select underneath, so it doesn't work properly if given a FD number >= 1024. This is a scalability bottleneck for hosts running large number of VMs.
To remove this limitation, we implemented a poll-based mechanism but with the same type signature as the Unix.select currently in use. So these two functions can be interchangeable at any stage.
--- tools/ocaml/xenstored/Makefile | 9 +++-- tools/ocaml/xenstored/select.ml | 54 ++++++++++++++++++++++++++++ tools/ocaml/xenstored/select.mli | 20 +++++++++++ tools/ocaml/xenstored/select_stubs.c | 68 ++++++++++++++++++++++++++++++++++++ tools/ocaml/xenstored/xenstored.ml | 2 +- 5 files changed, 149 insertions(+), 4 deletions(-) create mode 100644 tools/ocaml/xenstored/select.ml create mode 100644 tools/ocaml/xenstored/select.mli create mode 100644 tools/ocaml/xenstored/select_stubs.c
tools/ocaml/xenstored/Makefile +6 -3
@@ -15,10 +15,12 @@ OCAMLINCLUDE += \ -I $(OCAML_TOPLEVEL)/libs/xc \ -I $(OCAML_TOPLEVEL)/libs/eventchn -LIBS = syslog.cma syslog.cmxa +LIBS = syslog.cma syslog.cmxa select.cma select.cmxa syslog_OBJS = syslog syslog_C_OBJS = syslog_stubs -OCAML_LIBRARY = syslog +select_OBJS = select +select_C_OBJS = select_stubs +OCAML_LIBRARY = syslog select LIBS += systemd.cma systemd.cmxa systemd_OBJS = systemd
@@ -46,12 +48,13 @@ OBJS = define \ process \ xenstored -INTF = symbol.cmi trie.cmi syslog.cmi systemd.cmi +INTF = symbol.cmi trie.cmi syslog.cmi systemd.cmi select.cmi XENSTOREDLIBS = \ unix.cmxa \ -ccopt -L -ccopt . syslog.cmxa \ -ccopt -L -ccopt . systemd.cmxa \ + -ccopt -L -ccopt . select.cmxa \ -ccopt -L -ccopt $(OCAML_TOPLEVEL)/libs/mmap $(OCAML_TOPLEVEL)/libs/mmap/xenmmap.cmxa \ -ccopt -L -ccopt $(OCAML_TOPLEVEL)/libs/eventchn $(OCAML_TOPLEVEL)/libs/eventchn/xeneventchn.cmxa \ -ccopt -L -ccopt $(OCAML_TOPLEVEL)/libs/xc $(OCAML_TOPLEVEL)/libs/xc/xenctrl.cmxa \
tools/ocaml/xenstored/select.ml new file +54 -0
54 lines not on this page · raw message
tools/ocaml/xenstored/select.mli new file +20 -0
@@ -0,0 +1,20 @@ +(* + * Copyright (C) 2014 Zheng Li <[email protected]> + * + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU Lesser General Public License as published + * by the Free Software Foundation; version 2.1 only. with the special + * exception on linking described in file LICENSE. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU Lesser General Public License for more details. + *) + + +(** Same interface and semantics as [Unix.select] but with an extra alternative + implementation based on poll. *) +val select: + Unix.file_descr list -> Unix.file_descr list -> Unix.file_descr list -> float + -> Unix.file_descr list * Unix.file_descr list * Unix.file_descr list
tools/ocaml/xenstored/select_stubs.c new file +68 -0
68 lines not on this page · raw message
tools/ocaml/xenstored/xenstored.ml +1 -1
@@ -368,7 +368,7 @@ let _ = let timeout = if List.length mw > 0 then 0. else -1. in let rset, wset, _ = try - Unix.select (spec_fds @ inset) outset [] timeout + Select.select (spec_fds @ inset) outset [] timeout with Unix.Unix_error(Unix.EINTR, _, _) -> [], [], [] in let sfds, cfds =
Signature · 1 line
2.1.1
PATCH v3 2/9 oxenstored: add facilities to raise the max open fds uplimit
In reply to Zheng LiTo go beyond 1024 fds, we also need to raise the process limitation on max open fds (usually defaults to 1024).
We need to know the system level max open fds so that we won't go above that. Simply setting the limit to RLIM_INFINITY doesn't work on Linux 3.x (EPERM), a patch on this went into the 2.x branch but not 3.x for some reason.
--- tools/ocaml/xenstored/select.ml | 10 ++++++++++ tools/ocaml/xenstored/select_stubs.c | 12 ++++++++++++ 2 files changed, 22 insertions(+)
tools/ocaml/xenstored/select.ml +10 -0
@@ -23,6 +23,16 @@ type event = { } external select_on_poll: (Unix.file_descr * event) array -> int -> int = "stub_select_on_poll" +external set_fd_limit: int -> unit = "stub_set_fd_limit" + +(* The rlim_max given to setrlimit must not go above the system level nr_open, + which we can read from /proc/sys. *) +let get_sys_fs_nr_open () = + try + let ch = open_in "/proc/sys/fs/nr_open" in + let v = int_of_string (input_line ch) in + close_in_noerr ch; v + with _ -> 1024 * 1024 let init_event () = {read = false; write = false; except = false}
tools/ocaml/xenstored/select_stubs.c +12 -0
@@ -66,3 +66,15 @@ CAMLprim value stub_select_on_poll(value fd_events, value timeo) {CAMLreturn (Val_int (rc)); } + + +CAMLprim value stub_set_fd_limit(value limit) { + +CAMLparam1 (limit); +struct rlimit rl; + + rl.rlim_cur = rl.rlim_max =Int_val (limit); +if (setrlimit (RLIMIT_NOFILE , &rl) !=0 )uerror ("setrlimit" , Nothing); +CAMLreturn (Val_unit); + +}
Signature · 1 line
2.1.1
PATCH v3 3/9 oxenstored: add a --use-select command line flag
In reply to Zheng LiThis allows to fall back to the original Unix.select if preferred. It could be useful for debugging purposes too.
--- tools/ocaml/xenstored/parse_arg.ml | 8 ++++++-- tools/ocaml/xenstored/select.ml | 15 ++++++++++++++- tools/ocaml/xenstored/select.mli | 9 ++++++++- tools/ocaml/xenstored/xenstored.ml | 2 ++ 4 files changed, 30 insertions(+), 4 deletions(-)
tools/ocaml/xenstored/parse_arg.ml +6 -2
@@ -25,6 +25,7 @@ type config = tracefile: string option; (* old xenstored compatibility *) restart: bool; disable_socket: bool; + use_select: bool; } let do_argv =
@@ -35,7 +36,8 @@ let do_argv = and reraise_top_level = ref false and config_file = ref "" and restart = ref false - and disable_socket = ref false in + and disable_socket = ref false + and use_select = ref false in let speclist = [ ("--no-domain-init", Arg.Unit (fun () -> domain_init := false),
@@ -52,8 +54,9 @@ let do_argv = ("-T", Arg.Set_string tracefile, ""); (* for compatibility *) ("--restart", Arg.Set restart, "Read database on starting"); ("--disable-socket", Arg.Unit (fun () -> disable_socket := true), "Disable socket"); + ("--use-select", Arg.Unit (fun () -> use_select := true), "Use select instead of poll"); (* for backward compatibility and testing *) ] in - let usage_msg = "usage : xenstored [--config-file <filename>] [--no-domain-init] [--help] [--no-fork] [--reraise-top-level] [--restart] [--disable-socket]" in + let usage_msg = "usage : xenstored [--config-file <filename>] [--no-domain-init] [--help] [--no-fork] [--reraise-top-level] [--restart] [--disable-socket] [--use-select]" in Arg.parse speclist (fun s -> ()) usage_msg; { domain_init = !domain_init;
@@ -65,4 +68,5 @@ let do_argv = tracefile = if !tracefile <> "" then Some !tracefile else None; restart = !restart; disable_socket = !disable_socket; + use_select = !use_select; }
tools/ocaml/xenstored/select.ml +14 -1
@@ -36,7 +36,7 @@ let get_sys_fs_nr_open () = let init_event () = {read = false; write = false; except = false} -let select in_fds out_fds exc_fds timeout = +let poll_select in_fds out_fds exc_fds timeout = let h = Hashtbl.create 57 in let add_event event_set fd = let e =
@@ -62,3 +62,16 @@ let select in_fds out_fds exc_fds timeout = (if event.write then fd :: w else w), (if event.except then fd :: x else x)) a r + +(* If the use_poll function is not called at all, we default to the original Unix.select behavior *) +let select_fun = ref Unix.select + +let use_poll yes = + let sel_fun, max_fd = + if yes then poll_select, get_sys_fs_nr_open () + else Unix.select, 1024 in + select_fun := sel_fun; + set_fd_limit max_fd + +let select in_fds out_fds exc_fds timeout = + (!select_fun) in_fds out_fds exc_fds timeout
tools/ocaml/xenstored/select.mli +8 -1
@@ -14,7 +14,14 @@ (** Same interface and semantics as [Unix.select] but with an extra alternative - implementation based on poll. *) + implementation based on poll. Switching implementations is done by calling + the [use_poll] function. *) val select: Unix.file_descr list -> Unix.file_descr list -> Unix.file_descr list -> float -> Unix.file_descr list * Unix.file_descr list * Unix.file_descr list + +(** [use_poll true] will use poll based select with max fds number limitation + eliminated; [use_poll false] will use standard [Unix.select] with max fd + number set to 1024; not calling this function at all equals to use the + standard [Unix.select] with max fd number setting untouched. *) +val use_poll: bool -> unit
tools/ocaml/xenstored/xenstored.ml +2 -0
@@ -274,6 +274,8 @@ let _ = ); ); + Select.use_poll (not cf.use_select); + Sys.set_signal Sys.sighup (Sys.Signal_handle sighup_handler); Sys.set_signal Sys.sigterm (Sys.Signal_handle (fun i -> quit := true)); Sys.set_signal Sys.sigusr1 (Sys.Signal_handle (fun i -> sigusr1_handler store));
Signature · 1 line
2.1.1
PATCH v3 4/9 oxenstored: catch the error when a connection is already deleted
In reply to Zheng LiThe function process_fdset_with is called on the read set connections first. During the process, it might destroy a connection and remove it from the connections database if some errors occur. However, a reference to the same connection might still exist in the write set, which is awaiting to be processed next. In this case, a Not_found error will be raised and the process is aborted.
This patch changes the logic to ignore connections just missing from the connection database and continue the rest part of the work.
--- tools/ocaml/xenstored/xenstored.ml | 7 +++++-- 1 file changed, 5 insertions(+), 2 deletions(-)
tools/ocaml/xenstored/xenstored.ml +5 -2
@@ -43,8 +43,11 @@ let process_connection_fds store cons domains rset wset = debug "closing socket connection" in let process_fdset_with fds fct = - List.iter (fun fd -> try_fct fct (Connections.find cons fd)) fds - in + List.iter + (fun fd -> + try try_fct fct (Connections.find cons fd) + with Not_found -> () + ) fds in process_fdset_with rset Process.do_input; process_fdset_with wset Process.do_output
Signature · 1 line
2.1.1
PATCH v3 5/9 oxenstored: use hash table to store socket connections
In reply to Zheng LiCurrently we use list to store socket connections. This is fine for smaller number of connections. But when we scale up, traveling through a list of hundreds or thousands of connections just to find a single one of them is very low efficient.
This patch replaces the list with a (Unix.file_descr -> Connection.t) hash table.
--- tools/ocaml/xenstored/connections.ml | 28 ++++++++++++++-------------- 1 file changed, 14 insertions(+), 14 deletions(-)
tools/ocaml/xenstored/connections.ml +14 -14
@@ -18,17 +18,17 @@ let debug fmt = Logging.debug "connections" fmt type t = { - mutable anonymous: Connection.t list; + anonymous: (Unix.file_descr, Connection.t) Hashtbl.t; domains: (int, Connection.t) Hashtbl.t; mutable watches: (string, Connection.watch list) Trie.t; } -let create () = { anonymous = []; domains = Hashtbl.create 8; watches = Trie.create () } +let create () = { anonymous = Hashtbl.create 37; domains = Hashtbl.create 37; watches = Trie.create () } let add_anonymous cons fd can_write = let xbcon = Xenbus.Xb.open_fd fd in let con = Connection.create xbcon None in - cons.anonymous <- con :: cons.anonymous + Hashtbl.add cons.anonymous (Xenbus.Xb.get_fd xbcon) con let add_domain cons dom = let xbcon = Xenbus.Xb.open_mmap (Domain.get_interface dom) (fun () -> Domain.notify dom) in
@@ -36,14 +36,14 @@ let add_domain cons dom = Hashtbl.add cons.domains (Domain.get_id dom) con let select cons = - let inset = List.map (fun c -> Connection.get_fd c) cons.anonymous - and outset = List.fold_left (fun l c -> if Connection.has_output c - then Connection.get_fd c :: l - else l) [] cons.anonymous in - inset, outset + Hashtbl.fold + (fun _ con (ins, outs) -> + let fd = Connection.get_fd con in + (fd :: ins, if Connection.has_output con then fd :: outs else outs)) + cons.anonymous ([], []) -let find cons fd = - List.find (fun c -> Connection.get_fd c = fd) cons.anonymous +let find cons = + Hashtbl.find cons.anonymous let find_domain cons id = Hashtbl.find cons.domains id
@@ -55,7 +55,7 @@ let del_watches_of_con con watches = let del_anonymous cons con = try - cons.anonymous <- Utils.list_remove con cons.anonymous; + Hashtbl.remove cons.anonymous (Connection.get_fd con); cons.watches <- Trie.map (del_watches_of_con con) cons.watches; Connection.close con with exn ->
@@ -74,7 +74,7 @@ let iter_domains cons fct = Hashtbl.iter (fun k c -> fct c) cons.domains let iter_anonymous cons fct = - List.iter (fun c -> fct c) (List.rev cons.anonymous) + Hashtbl.iter (fun _ c -> fct c) cons.anonymous let iter cons fct = iter_domains cons fct; iter_anonymous cons fct
@@ -163,10 +163,10 @@ let stats cons = nb_ops_dom := !nb_ops_dom + con_ops; nb_watchs_dom := !nb_watchs_dom + con_watchs; ); - (List.length cons.anonymous, !nb_ops_anon, !nb_watchs_anon, + (Hashtbl.length cons.anonymous, !nb_ops_anon, !nb_watchs_anon, Hashtbl.length cons.domains, !nb_ops_dom, !nb_watchs_dom) let debug cons = - let anonymous = List.map Connection.debug cons.anonymous in + let anonymous = Hashtbl.fold (fun _ con accu -> Connection.debug con :: accu) cons.anonymous [] in let domains = Hashtbl.fold (fun _ con accu -> Connection.debug con :: accu) cons.domains [] in String.concat "" (domains @ anonymous)
Signature · 1 line
2.1.1
PATCH v3 6/9 oxenstored: enable domain connection indexing based on eventchn port
In reply to Zheng LiCurrently in xenstore connection database, we use a hash table of (domid -> connection) to store domain connections. This allows fast indexing based on dom ids.
This patch adds another dimention of fast indexing that is based on eventchn port number. This is useful when doing selective connection processing based on the port numbers of incoming events.
--- tools/ocaml/xenstored/connections.ml | 26 ++++++++++++++++++++++---- tools/ocaml/xenstored/domain.ml | 1 + 2 files changed, 23 insertions(+), 4 deletions(-)
tools/ocaml/xenstored/connections.ml +22 -4
@@ -20,10 +20,16 @@ let debug fmt = Logging.debug "connections" fmt type t = { anonymous: (Unix.file_descr, Connection.t) Hashtbl.t; domains: (int, Connection.t) Hashtbl.t; + ports: (Xeneventchn.t, Connection.t) Hashtbl.t; mutable watches: (string, Connection.watch list) Trie.t; } -let create () = { anonymous = Hashtbl.create 37; domains = Hashtbl.create 37; watches = Trie.create () } +let create () = { + anonymous = Hashtbl.create 37; + domains = Hashtbl.create 37; + ports = Hashtbl.create 37; + watches = Trie.create () +} let add_anonymous cons fd can_write = let xbcon = Xenbus.Xb.open_fd fd in
@@ -33,7 +39,10 @@ let add_anonymous cons fd can_write = let add_domain cons dom = let xbcon = Xenbus.Xb.open_mmap (Domain.get_interface dom) (fun () -> Domain.notify dom) in let con = Connection.create xbcon (Some dom) in - Hashtbl.add cons.domains (Domain.get_id dom) con + Hashtbl.add cons.domains (Domain.get_id dom) con; + match Domain.get_port dom with + | Some p -> Hashtbl.add cons.ports p con; + | None -> () let select cons = Hashtbl.fold
@@ -45,8 +54,11 @@ let select cons = let find cons = Hashtbl.find cons.anonymous -let find_domain cons id = - Hashtbl.find cons.domains id +let find_domain cons = + Hashtbl.find cons.domains + +let find_domain_by_port cons port = + Hashtbl.find cons.ports port let del_watches_of_con con watches = match List.filter (fun w -> Connection.get_con w != con) watches with
@@ -65,6 +77,12 @@ let del_domain cons id = try let con = find_domain cons id in Hashtbl.remove cons.domains id; + (match Connection.get_domain con with + | Some d -> + (match Domain.get_port d with + | Some p -> Hashtbl.remove cons.ports p + | None -> ()) + | None -> ()); cons.watches <- Trie.map (del_watches_of_con con) cons.watches; Connection.close con with exn ->
tools/ocaml/xenstored/domain.ml +1 -0
@@ -35,6 +35,7 @@ let get_id domain = domain.id let get_interface d = d.interface let get_mfn d = d.mfn let get_remote_port d = d.remote_port +let get_port d = d.port let is_bad_domain domain = domain.bad_client let mark_as_bad domain = domain.bad_client <- true
Signature · 1 line
2.1.1
PATCH v3 7/9 oxenstored: only process domain connections that notify us by events
In reply to Zheng LiCurrently, upon receiving an event, oxenstored will always scan/process all the domain connections (xs rings), disregarding which domain sent that event. This is rather costy and inefficient. It also shadows and indulges client for not correctly communicating with us on message/space availability.
With this patch, oxenstore will only scan/process the domain connections that have correctly notified us by events or have IO actions leftover from previous communication.
--- tools/ocaml/xenstored/connection.ml | 4 ++++ tools/ocaml/xenstored/connections.ml | 9 ++++----- tools/ocaml/xenstored/xenstored.ml | 13 ++++++++++--- 3 files changed, 18 insertions(+), 8 deletions(-)
tools/ocaml/xenstored/connection.ml +4 -0
@@ -223,10 +223,14 @@ let pop_in con = Xenbus.Xb.get_in_packet con.xb let has_more_input con = Xenbus.Xb.has_more_input con.xb let has_output con = Xenbus.Xb.has_output con.xb +let has_old_output con = Xenbus.Xb.has_old_output con.xb let has_new_output con = Xenbus.Xb.has_new_output con.xb let peek_output con = Xenbus.Xb.peek_output con.xb let do_output con = Xenbus.Xb.output con.xb +let has_more_work con = + has_more_input con || not (has_old_output con) && has_new_output con + let incr_ops con = con.stat_nb_ops <- con.stat_nb_ops + 1 let mark_symbols con =
tools/ocaml/xenstored/connections.ml +4 -5
@@ -98,11 +98,10 @@ let iter cons fct = iter_domains cons fct; iter_anonymous cons fct let has_more_work cons = - Hashtbl.fold (fun id con acc -> - if Connection.has_more_input con then - con :: acc - else - acc) cons.domains [] + Hashtbl.fold + (fun id con acc -> + if Connection.has_more_work con then con :: acc else acc) + cons.domains [] let key_of_str path = if path.[0] = '@'
tools/ocaml/xenstored/xenstored.ml +10 -3
@@ -57,7 +57,10 @@ let process_domains store cons domains = let con = Connections.find_domain cons (Domain.get_id domain) in Process.do_input store cons domains con; Process.do_output store cons domains con in - Domains.iter domains do_io_domain + List.iter + (fun c -> + match Connection.get_domain c with + | Some d -> do_io_domain d | _ -> ()) let sigusr1_handler store = try
@@ -305,6 +308,7 @@ let _ = Connections.add_anonymous cons cfd can_write and handle_eventchn fd = let port = Event.pending eventchn in + debug "pending port %d" (Xeneventchn.to_int port); finally (fun () -> if Some port = eventchn.Event.virq_port then ( let (notify, deaddom) = Domains.cleanup xc domains in
@@ -312,7 +316,10 @@ let _ = if deaddom <> [] || notify then Connections.fire_spec_watches cons "@releaseDomain" ) - ) (fun () -> Event.unmask eventchn port); + else + let c = Connections.find_domain_by_port cons port in + process_domains store cons domains [c] + ) (fun () -> Event.unmask eventchn port) and do_if_set fd set fct = if List.mem fd set then fct fd in
@@ -382,7 +389,7 @@ let _ = process_special_fds sfds; if List.length cfds > 0 || List.length wset > 0 then process_connection_fds store cons domains cfds wset; - process_domains store cons domains + process_domains store cons domains mw in while not !quit
Signature · 1 line
2.1.1
PATCH v3 8/9 oxenstored: add a safe net mechanism for existing ill-behaved clients
In reply to Zheng LiIn previous commit, we moved from exhaustively scanning all domain connections to only processing those have correctly notified us by events. The benefits are not only efficiency but also correctness, because it could potentially block an ill-behaved client and have it waiting on its own mistake. If someone makes a mistake on this when developing a piece of code, he/she would immediately notice the problem (as the process being blocked), so that he/she could fix it rightaway before anything else. Note that the chances of making such mistakes are rare in reality, because most client code would use the libxenstore library (which has all the notification logic built in correctly) instead of having to implement raw accessing from scratch.
On the other hand, we did notice that there were some legacy code that didn't do the notification correctly. As some code might be still running in wild, it would be bad if they break by this change (e.g. after an upgrade). This patch introduces a safe net mechanism to ensure ill-behaved clients continue to work, but still retain most of the performance benefits here.
* We add a checker to still scan all the rings periodically, so that we can still pick up these messages at an acceptable frequency.
* Internally, we introduce an io_credit concept for domain connections. It represents the rounds of ring scan we are going to perform on a domain connection. For well-behaved connections, this value is changing between 0 and 1; but for connections detected as ill-behaved, we'll bump its credit to a high value so that we'll unconditionally scan its ring for the next $n$ rounds. This way, the client won't hiccupped by the interval between checker's running (especially during periods when it continously interacts with oxenstored); and oxenstored doesn't have to keep scanning these rings indefinitely (with the credit running out), as they are usually quite most of the time.
* We log an message when a domain connection is suspected as ill-behaved. Enable [info] level logging if you want/need to see it in action. Note that this information won't be accurate, as false positives are possible due to time window (e.g. we detect a client has written to the ring and we get no notificiation from it for the time being, but still the notification could potentially arrive at some time later). It's no harm to give a domain connection extra credit though.
--- tools/ocaml/xenstored/domain.ml | 11 ++++- tools/ocaml/xenstored/oxenstored.conf | 3 ++ tools/ocaml/xenstored/xenstored.ml | 76 ++++++++++++++++++++++++++--------- 3 files changed, 69 insertions(+), 21 deletions(-)
tools/ocaml/xenstored/domain.ml +10 -1
@@ -28,6 +28,9 @@ type t = eventchn: Event.t; mutable port: Xeneventchn.t option; mutable bad_client: bool; + mutable io_credit: int; (* the rounds of ring process left to do, default is 0, + usually set to 1 when there is work detected, could + also set to n to give "lazy" clients extra credit *) } let get_path dom = "/local/domain/" ^ (sprintf "%u" dom.id)
@@ -40,6 +43,11 @@ let get_port d = d.port let is_bad_domain domain = domain.bad_client let mark_as_bad domain = domain.bad_client <- true +let get_io_credit domain = domain.io_credit +let set_io_credit ?(n=1) domain = domain.io_credit <- max 0 n +let incr_io_credit domain = domain.io_credit <- domain.io_credit + 1 +let decr_io_credit domain = domain.io_credit <- max 0 (domain.io_credit - 1) + let string_of_port = function | None -> "None" | Some x -> string_of_int (Xeneventchn.to_int x)
@@ -74,7 +82,8 @@ let make id mfn remote_port interface eventchn = { interface = interface; eventchn = eventchn; port = None; - bad_client = false + bad_client = false; + io_credit = 0; } let is_dom0 d = d.id = 0
tools/ocaml/xenstored/oxenstored.conf +3 -0
@@ -33,3 +33,6 @@ persistent = false # acesss-log-nb-chars = 180 # access-log-special-ops = false +# Perodically scanning all the rings as a safenet for lazy clients. +# Define the interval in seconds, set to negative to disable. +# ring-scan-interval = 20
tools/ocaml/xenstored/xenstored.ml +56 -20
132 lines not on this page · raw message
Signature · 1 line
2.1.1
PATCH v3 9/9 oxenstored: reduce syslog call overhead
In reply to Zheng LiWe noticed that, if configured to use syslog as the logging backend, every single line of access logging (via the syslog C binding) will call stat on /etc/localtime for 3 times. The rational behind this is probably to detect any timezone changes over time.
This is a considerable cost we'd like to avoid, given the intensiveness of our access logging --- we log almost every xenstore status change (for good reason). Also a running Xen host is rarely a mobile environment, so the little benefit can hardly justify the cost.
Setting up the TZ environment varialbe can avoid stat calls.
--- tools/ocaml/xenstored/logging.ml | 11 ++++++++++- 1 file changed, 10 insertions(+), 1 deletion(-)
tools/ocaml/xenstored/logging.ml +10 -1
@@ -130,8 +130,17 @@ let string_of_date () = tm.Unix.tm_hour tm.Unix.tm_min tm.Unix.tm_sec (int_of_float (1000.0 *. msec)) +(* We can defer to syslog for log management *) let make_syslog_logger facility = - (* We defer to syslog for log management *) + (* When TZ is unset in the environment, each syslog call will stat the + /etc/localtime file at least three times during the process. We'd like to + avoid this cost given that we are not a mobile environment and we log + almost every xenstore entry update/watch. *) + let () = + let tz_is_set = + try String.length (Unix.getenv "TZ") > 0 + with Not_found -> false in + if not tz_is_set then Unix.putenv "TZ" "/etc/localtime" in let nothing () = () in let write ?level s = let level = match level with
Signature · 1 line
2.1.1
Re: [PATCH v3 9/9] oxenstored: reduce syslog call overhead (fix typo)
We noticed that, if configured to use syslog as the logging backend, every single line of access logging (via the syslog C binding) will call stat on /etc/localtime for 3 times. The rational behind this is probably to detect any timezone changes over time.
This is a considerable cost we'd like to avoid, given the intensiveness of our access logging --- we log almost every xenstore status change (for good reason). Also a running Xen host is rarely a mobile environment, so the little benefit can hardly justify the cost.
Setting up the TZ environment varialbe to avoid stat calls.
--- tools/ocaml/xenstored/logging.ml | 11 ++++++++++- 1 file changed, 10 insertions(+), 1 deletion(-)
tools/ocaml/xenstored/logging.ml +10 -1
@@ -130,8 +130,17 @@ let string_of_date () = tm.Unix.tm_hour tm.Unix.tm_min tm.Unix.tm_sec (int_of_float (1000.0 *. msec)) +(* We can defer to syslog for log management *) let make_syslog_logger facility = - (* We defer to syslog for log management *) + (* When TZ is unset in the environment, each syslog call will stat the + /etc/localtime file three times during the process. We'd like to + avoid this cost given that we are not a mobile environment and we log + almost every xenstore entry update/watch. *) + let () = + let tz_is_set = + try String.length (Unix.getenv "TZ") > 0 + with Not_found -> false in + if not tz_is_set then Unix.putenv "TZ" ":/etc/localtime" in let nothing () = () in let write ?level s = let level = match level with
Signature · 1 line
2.1.1
Re: [PATCH v3 9/9] oxenstored: reduce syslog call overhead (fix typo)
Zheng Li writes ("Re: [PATCH v3 9/9] oxenstored: reduce syslog call overhead (fix typo)"):
Quoted 1 line
Setting up the TZ environment varialbe to avoid stat calls.
...
tools/ocaml/xenstored/logging.ml let string_of_date () = · line 143
Quoted 1 line
tools/ocaml/xenstored/logging.ml +1
-0+ if not tz_is_set then Unix.putenv "TZ" ":/etc/localtime" in
I wasn't aware of this syntax. Can you point me to its specification ?
Thanks,
Ian.
Re: [PATCH v3 9/9] oxenstored: reduce syslog call overhead (fix typo)
Hi Ian,
Quoted 6 lines · Ian Jackson
Zheng Li writes ("Re: [PATCH v3 9/9] oxenstored: reduce syslog call overhead (fix typo)"):
Quoted 1 line
Setting up the TZ environment varialbe to avoid stat calls.
...
tools/ocaml/xenstored/logging.ml let string_of_date () = · line 143
Quoted 1 line
tools/ocaml/xenstored/logging.ml +1
-0+ if not tz_is_set then Unix.putenv "TZ" ":/etc/localtime" inI wasn't aware of this syntax. Can you point me to its specification ?
Please see the 3rd format (:characters format) in the linked document below
http://www.gnu.org/software/libc/manual/html_node/TZ-Variable.html
Also a similar issue reported on stackoverflow and discussion
I've manually tested the change. With TZ set, strace no longer shows the intensive stat calls, and the timestamps seem to be correct wrt the local time setting.
Cheers,
Zheng
Re: [PATCH v3 9/9] oxenstored: reduce syslog call overhead (fix typo)
Zheng Li writes ("Re: [PATCH v3 9/9] oxenstored: reduce syslog call overhead (fix typo)"):
Quoted 2 lines
Please see the 3rd format (:characters format) in the linked document below
http://www.gnu.org/software/libc/manual/html_node/TZ-Variable.html
That's great for glibc-based guests. But what about the BSDs ?
I looked at
http://pubs.opengroup.org/onlinepubs/9699919799/basedefs/V1_chap08.html#tag_08
and it just says that the : syntax is `implementation defined'.
Thanks,
Ian.
Re: [PATCH v3 9/9] oxenstored: reduce syslog call overhead (fix typo)
Quoted 5 lines · Ian Jackson
Zheng Li writes ("Re: [PATCH v3 9/9] oxenstored: reduce syslog call overhead (fix typo)"):
Quoted 2 lines
Please see the 3rd format (:characters format) in the linked document below
http://www.gnu.org/software/libc/manual/html_node/TZ-Variable.htmlThat's great for glibc-based guests. But what about the BSDs ?
This code does xenstored access logging in Dom0. Looking at the supported Dom0 distros page on the wiki, its seems that only NetBSD is the concern here?
Quoted 3 lines
I looked at
http://pubs.opengroup.org/onlinepubs/9699919799/basedefs/V1_chap08.html#tag_08
and it just says that the : syntax is `implementation defined'.
I guess most BSDs adopt the same semantics. I searched for NetBSD libc manual and found this (starting from "if the value does not begin with a colon ...")
http://netbsd.gw.com/cgi-bin/man-cgi?tzset+3
Cheers,
Zheng
Re: [PATCH v3 9/9] oxenstored: reduce syslog call overhead (fix typo)
Zheng Li writes ("Re: [PATCH v3 9/9] oxenstored: reduce syslog call overhead (fix typo)"):
Quoted 8 lines · Ian Jackson
Quoted 5 lines
Zheng Li writes ("Re: [PATCH v3 9/9] oxenstored: reduce syslog call overhead (fix typo)"):
Quoted 2 lines
Please see the 3rd format (:characters format) in the linked document below
http://www.gnu.org/software/libc/manual/html_node/TZ-Variable.htmlThat's great for glibc-based guests. But what about the BSDs ?
This code does xenstored access logging in Dom0. Looking at the supported Dom0 distros page on the wiki, its seems that only NetBSD is the concern here?
Indeed, although we might expect to port to FreeBSD too.
Quoted 7 lines
Quoted 3 lines
I looked at
http://pubs.opengroup.org/onlinepubs/9699919799/basedefs/V1_chap08.html#tag_08
and it just says that the : syntax is `implementation defined'.I guess most BSDs adopt the same semantics. I searched for NetBSD libc manual and found this (starting from "if the value does not begin with a colon ...")
Ah, good. I doubt NetBSD has diverged much here, so that's OK.
Thanks,
Ian.
Re: [PATCH v3 9/9] oxenstored: reduce syslog call overhead (fix typo)
In reply to Zheng LiQuoted 7 lines · Zheng Li
Quoted 5 lines · Ian Jackson
Zheng Li writes ("Re: [PATCH v3 9/9] oxenstored: reduce syslog call overhead (fix typo)"):
Quoted 2 lines
Please see the 3rd format (:characters format) in the linked document below
http://www.gnu.org/software/libc/manual/html_node/TZ-Variable.htmlThat's great for glibc-based guests. But what about the BSDs ?
Quoted 7 lines
This code does xenstored access logging in Dom0. Looking at the supported Dom0 distros page on the wiki, its seems that only NetBSD is the concern here?
Quoted 3 lines
I looked at
http://pubs.opengroup.org/onlinepubs/9699919799/basedefs/V1_chap08.html#tag_08
and it just says that the : syntax is `implementation defined'.I guess most BSDs adopt the same semantics. I searched for NetBSD libc manual and found this (starting from "if the value does not begin with a colon ...")
Sorry, should be "if the value begins with a colon ...". From what I read, even if it doesn't start with a colon, it will still be treated as a time conversion file as the first resort.
Also, the syslog implementation on BSD might be different (e.g. might not have the "stat" issue at all). So as far as setting the TZ variable won't cause any problem for the xenstored process, we should be fine with it.
Cheers,
Zheng
Quoted 9 lines · Zheng Li
This is mainly about removing the 1024 fds limitation in the current oxenstored implementation. We also fixed some bugs and made some perf improvements along the way.
This is v3. The first 6 patches are the same as in v2. For patch 7/8/9, we
* Add a safe net mechanism for ill-behaved legacy clients
* Make some performance improvement to reduce syslog workload
* Small refactoring on patch 7/8 to accomodate the new changes
That all looks quite nice but I have no experience with OCaml.
I fear that this patchset will have wait until an OCaml expert can review the code :-(
Quoted 9 lines
Cheers,
Zheng_______________________________________________
Xen-devel mailing list
[email protected]
http://lists.xen.org/xen-devel
Dave Scott
Acked-byHi,
Quoted 7 lines · Konrad Rzeszutek Wilk
Quoted 6 lines · Zheng Li
This is mainly about removing the 1024 fds limitation in the current oxenstored implementation. We also fixed some bugs and made some perf improvements along the way.
This is v3. The first 6 patches are the same as in v2. For patch 7/8/9, we
Quoted 8 lines
Quoted 3 lines
* Add a safe net mechanism for ill-behaved legacy clients
* Make some performance improvement to reduce syslog workload
* Small refactoring on patch 7/8 to accomodate the new changesThat all looks quite nice but I have no experience with OCaml.
I fear that this patchset will have wait until an OCaml expert can review the code :-(
I’ve just read through the v3 patches and am happy with them. So I’m happy to give them an
I think this patch:
[PATCH v3 4/9] oxenstored: catch the error when a connection is already deleted
is a useful bug fix — this might explain why oxenstored has occasionally ‘disappeared’ in mysterious circumstances. Zheng: could this patch be taken by itself? If so I think the release should definitely have it.
The rest of the patches will help scalability a lot but aren’t critical fixes. They would (IMHO) increase the quality of the release though.
Cheers,
Dave
Quoted 14 lines
Quoted 9 lines
Cheers,
Zheng_______________________________________________
Xen-devel mailing list
[email protected]
http://lists.xen.org/xen-devel_______________________________________________
Xen-devel mailing list
[email protected]
http://lists.xen.org/xen-devel
Quoted 9 lines · Dave Scott
I’ve just read through the v3 patches and am happy with them. So I’m happy to give them an
Acked-by David Scott <[email protected]>I think this patch:
[PATCH v3 4/9] oxenstored: catch the error when a connection is already deleted
is a useful bug fix — this might explain why oxenstored has occasionally ‘disappeared’ in mysterious circumstances. Zheng: could this patch be taken by itself? If so I think the release should definitely have it.
Yes, [4/9] can be picked up by itself. It's a small patch and should be quite safe.
Quoted 1 line
The rest of the patches will help scalability a lot but aren’t critical fixes. They would (IMHO) increase the quality of the release though.
To group the these patches:
* [1/9], [2/9], and [3/9], are about select -> poll;
* [7/9] and [8/9] are about exhaustive ring scan -> selective ring scan, with [5/9] and [6/9] adding some facilities.
* [4/9] is a solo bug fix patch as stated;
* [9/9] is a solo perf improvement patch on syslog load.
Cheers,
Zheng
_______________________________________________
Xen-devel mailing list
[email protected]
http://lists.xen.org/xen-devel
Quoted 15 lines · Dave Scott
Hi,
Quoted 11 lines · Konrad Rzeszutek Wilk
Quoted 9 lines · Zheng Li
This is mainly about removing the 1024 fds limitation in the current oxenstored implementation. We also fixed some bugs and made some perf improvements along the way.
This is v3. The first 6 patches are the same as in v2. For patch 7/8/9, we
* Add a safe net mechanism for ill-behaved legacy clients
* Make some performance improvement to reduce syslog workload
* Small refactoring on patch 7/8 to accomodate the new changes
Quoted 8 lines
Quoted 4 lines
That all looks quite nice but I have no experience with OCaml.
I fear that this patchset will have wait until an OCaml expert can review the code :-(
I’ve just read through the v3 patches and am happy with them. So I’m happy to give them an
Acked-by David Scott <[email protected]>
Excellent!
Quoted 8 lines
I think this patch:
[PATCH v3 4/9] oxenstored: catch the error when a connection is already deleted
is a useful bug fix — this might explain why oxenstored has occasionally ‘disappeared’ in mysterious circumstances. Zheng: could this patch be taken by itself? If so I think the release should definitely have it.
The rest of the patches will help scalability a lot but aren’t critical fixes. They would (IMHO) increase the quality of the release though.
The bug-fix I believe should go in Xen 4.5.
In regards to the rest (help scalability) I am worried that:
- This is rather unknown code for me (OCaml) but then so is the ARM assembler (I am slowly digging through the manual) - so I am falling back here on the maintainers perspective. Dave says OK, so that means we are OK.
- Now on the other hand this is core code. If this goes belly up we a screwed - and if this introduces races, it will be quite a headache to troubleshoot down to this.
In terms of the positive aspects:
- It will improve the code quality.
- It improves scalability.
Dave, when you said "am happy with them" (and giving it an Acked-by) - does that mean you dug through the code carefully with a microscope looking for ways to break it? Or was it more of a cursory view?
Asking because if it is the first then an 'Reviewed-by' is more apt and it also means that the chance of the code going 'belly-up' is lesser.
If you are comfortable giving an 'Reviewed-by' then I believe it can go in Xen 4.5 (and lowers the 'belly-up' chance). If you don't have the cycles then I believe it should go in Xen 4.6 to give it some "soak" time.
If there are any doubts or if the author is not too keen on debugging issues during the next couple of months (say, if you have vacation planned or such) - if of course there are bugs - we can also postpone these patches and put them in Xen 4.6 and have a more time to sort issues out.
Thank you!
_______________________________________________
Xen-devel mailing list
[email protected]
http://lists.xen.org/xen-devel
Dave Scott
Reviewed-byHi,
Quoted 55 lines · Konrad Rzeszutek Wilk
Quoted 15 lines · Dave Scott
Hi,
Quoted 11 lines · Konrad Rzeszutek Wilk
Quoted 9 lines · Zheng Li
This is mainly about removing the 1024 fds limitation in the current oxenstored implementation. We also fixed some bugs and made some perf improvements along the way.
This is v3. The first 6 patches are the same as in v2. For patch 7/8/9, we
* Add a safe net mechanism for ill-behaved legacy clients
* Make some performance improvement to reduce syslog workload
* Small refactoring on patch 7/8 to accomodate the new changesQuoted 8 lines
Quoted 4 lines
That all looks quite nice but I have no experience with OCaml.
I fear that this patchset will have wait until an OCaml expert can review the code :-(
I’ve just read through the v3 patches and am happy with them. So I’m happy to give them an
Acked-by David Scott <[email protected]>Excellent!
Quoted 8 lines
I think this patch:
[PATCH v3 4/9] oxenstored: catch the error when a connection is already deleted
is a useful bug fix — this might explain why oxenstored has occasionally ‘disappeared’ in mysterious circumstances. Zheng: could this patch be taken by itself? If so I think the release should definitely have it.
The rest of the patches will help scalability a lot but aren’t critical fixes. They would (IMHO) increase the quality of the release though.
The bug-fix I believe should go in Xen 4.5.
In regards to the rest (help scalability) I am worried that:
- This is rather unknown code for me (OCaml) but then so is the ARM assembler (I am slowly digging through the manual) - so I am falling back here on the maintainers perspective. Dave says OK, so that means we are OK.
- Now on the other hand this is core code. If this goes belly up we a screwed - and if this introduces races, it will be quite a headache to troubleshoot down to this.In terms of the positive aspects:
- It will improve the code quality.
- It improves scalability.Dave, when you said "am happy with them" (and giving it an Acked-by) - does that mean you dug through the code carefully with a microscope looking for ways to break it? Or was it more of a cursory view?
Quoted 7 lines
Asking because if it is the first then an 'Reviewed-by' is more apt and it also means that the chance of the code going 'belly-up' is lesser.
If you are comfortable giving an 'Reviewed-by' then I believe it can go in Xen 4.5 (and lowers the 'belly-up' chance). If you don't have the cycles then I believe it should go in Xen 4.6 to give it some "soak" time.
I’ve re-examined the patches and can’t see anything wrong with them. I talked to Zheng off-list about how the patches were tested. Zheng has run a lot of test sequences, many of them are very xenstore-intensive. Reassuringly he found a bug — not in his patches, but in en earlier version of my ‘xenstore reconnection’ patch (subsequently fixed).
So I’m confident to say:
Cheers,
Dave
Quoted 13 lines
If there are any doubts or if the author is not too keen on debugging issues during the next couple of months (say, if you have vacation planned or such) - if of course there are bugs - we can also postpone these patches and put them in Xen 4.6 and have a more time to sort issues out.
Thank you!
_______________________________________________
Xen-devel mailing list
[email protected]
http://lists.xen.org/xen-devel
Quoted 73 lines · Dave Scott
Hi,
Quoted 55 lines · Konrad Rzeszutek Wilk
Quoted 15 lines · Dave Scott
Hi,
Quoted 11 lines · Konrad Rzeszutek Wilk
Quoted 9 lines · Zheng Li
This is mainly about removing the 1024 fds limitation in the current oxenstored implementation. We also fixed some bugs and made some perf improvements along the way.
This is v3. The first 6 patches are the same as in v2. For patch 7/8/9, we
* Add a safe net mechanism for ill-behaved legacy clients
* Make some performance improvement to reduce syslog workload
* Small refactoring on patch 7/8 to accomodate the new changesQuoted 8 lines
Quoted 4 lines
That all looks quite nice but I have no experience with OCaml.
I fear that this patchset will have wait until an OCaml expert can review the code :-(
I’ve just read through the v3 patches and am happy with them. So I’m happy to give them an
Acked-by David Scott <[email protected]>Excellent!
Quoted 8 lines
I think this patch:
[PATCH v3 4/9] oxenstored: catch the error when a connection is already deleted
is a useful bug fix — this might explain why oxenstored has occasionally ‘disappeared’ in mysterious circumstances. Zheng: could this patch be taken by itself? If so I think the release should definitely have it.
The rest of the patches will help scalability a lot but aren’t critical fixes. They would (IMHO) increase the quality of the release though.
The bug-fix I believe should go in Xen 4.5.
In regards to the rest (help scalability) I am worried that:
- This is rather unknown code for me (OCaml) but then so is the ARM assembler (I am slowly digging through the manual) - so I am falling back here on the maintainers perspective. Dave says OK, so that means we are OK.
- Now on the other hand this is core code. If this goes belly up we a screwed - and if this introduces races, it will be quite a headache to troubleshoot down to this.In terms of the positive aspects:
- It will improve the code quality.
- It improves scalability.Dave, when you said "am happy with them" (and giving it an Acked-by) - does that mean you dug through the code carefully with a microscope looking for ways to break it? Or was it more of a cursory view?
Quoted 7 lines
Asking because if it is the first then an 'Reviewed-by' is more apt and it also means that the chance of the code going 'belly-up' is lesser.
If you are comfortable giving an 'Reviewed-by' then I believe it can go in Xen 4.5 (and lowers the 'belly-up' chance). If you don't have the cycles then I believe it should go in Xen 4.6 to give it some "soak" time.
I’ve re-examined the patches and can’t see anything wrong with them. I talked to Zheng off-list about how the patches were tested. Zheng has run a lot of test sequences, many of them are very xenstore-intensive. Reassuringly he found a bug — not in his patches, but in en earlier version of my ‘xenstore reconnection’ patch (subsequently fixed).
Quoted 3 lines
So I’m confident to say:
Reviewed-by David Scott <[email protected]>
Excellent, hence:
Thank you!
Quoted 19 lines
Cheers,
DaveQuoted 13 lines
If there are any doubts or if the author is not too keen on debugging issues during the next couple of months (say, if you have vacation planned or such) - if of course there are bugs - we can also postpone these patches and put them in Xen 4.6 and have a more time to sort issues out.
Thank you!
_______________________________________________
Xen-devel mailing list
[email protected]
http://lists.xen.org/xen-devel
_______________________________________________
Xen-devel mailing list
[email protected]
http://lists.xen.org/xen-devel
Quoted 8 lines · Konrad Rzeszutek Wilk
Quoted 4 lines
So I’m confident to say:
Reviewed-by David Scott <[email protected]>Excellent, hence:
Release-Acked-by Konrad Rzeszutek Wilk <[email protected]>
I've applied all of them.
Including the final patch "oxenstored: reduce syslog call overhead (fix typo)", there was some back and forth on that one but Ian J confirmed over IRC that he was satisfied.
Quoted 22 lines
Thank you!
Quoted 19 lines
Cheers,
DaveQuoted 13 lines
If there are any doubts or if the author is not too keen on debugging issues during the next couple of months (say, if you have vacation planned or such) - if of course there are bugs - we can also postpone these patches and put them in Xen 4.6 and have a more time to sort issues out.
Thank you!
_______________________________________________
Xen-devel mailing list
[email protected]
http://lists.xen.org/xen-devel
Quoted 4 lines
_______________________________________________
Xen-devel mailing list
[email protected]
http://lists.xen.org/xen-devel
_______________________________________________
Xen-devel mailing list
[email protected]
http://lists.xen.org/xen-devel