[Xen-API] [PATCH 2 of 2] Add RRDs to measure the use of the local disk cache

[Jon Ludlam <jonathan.ludlam@xxxxxxxxxxxxx>]

# HG changeset patch
# User Jon Ludlam <Jonathan.Ludlam@xxxxxxxxxxxxx>
# Date 1285345651 -3600
# Node ID c9096a7ecb48a10462c280b8b672e8bf04bbd3e1
# Parent  243ef81d48f72655dc409b127eb39785e4a5f5c0
Add RRDs to measure the use of the local disk cache.

Signed-off-by: Jon Ludlam <Jonathan.Ludlam@xxxxxxxxxxxxx>

diff -r 243ef81d48f7 -r c9096a7ecb48 ocaml/xapi/dbsync_slave.ml
--- a/ocaml/xapi/dbsync_slave.ml
+++ b/ocaml/xapi/dbsync_slave.ml
@@ -480,6 +480,15 @@
   (* record who we are in xapi_globs *)
   Xapi_globs.localhost_ref := Helpers.get_localhost ~__context;
 
+  (* Set the cache_sr *)
+  begin 
+         try
+                 let cache_sr = Db.Host.get_local_cache_sr ~__context 
~self:(Helpers.get_localhost ~__context) in
+                 let cache_sr_uuid = Db.SR.get_uuid ~__context ~self:cache_sr 
in
+                 Monitor.set_cache_sr cache_sr_uuid
+         with _ -> Monitor.unset_cache_sr () 
+  end;
+  
   (* Load the host rrd *)
   Monitor_rrds.load_rrd ~__context (Helpers.get_localhost_uuid ()) true;
 
diff -r 243ef81d48f7 -r c9096a7ecb48 ocaml/xapi/monitor.ml
--- a/ocaml/xapi/monitor.ml
+++ b/ocaml/xapi/monitor.ml
@@ -422,6 +422,77 @@
    (Host,ds_make ~name:"xapi_allocation_kib" ~description:"Memory allocation 
done by the xapi daemon"
      ~value:(Rrd.VT_Float xapigrad_kib) ~ty:Rrd.Derive ~min:0.0 ~default:true 
());
   ]
+
+(**** Local cache SR stuff *)
+
+type last_vals = {
+       time : float;
+       cache_size_raw : int64;
+       cache_hits_raw : int64;
+       cache_misses_raw : int64;
+}
+
+let last_cache_stats = ref None
+let cache_sr_uuid = ref None
+let cache_sr_lock = Mutex.create () 
+let cached_cache_dss = ref []
+
+(* Avoid a db lookup every 5 secs by caching the cache sr uuid. Set by 
xapi_host and dbsync_slave *)
+let set_cache_sr sr_uuid = 
+       Mutex.execute cache_sr_lock (fun () -> cache_sr_uuid := Some sr_uuid)
+
+let unset_cache_sr () =
+       Mutex.execute cache_sr_lock (fun () -> cache_sr_uuid := None)
+
+let read_cache_stats timestamp =
+       let cache_sr_opt = Mutex.execute cache_sr_lock (fun () -> 
!cache_sr_uuid) in
+
+       let do_read cache_sr =
+               let (cache_stats_out,err) = 
Forkhelpers.execute_command_get_output "/opt/xensource/bin/tapdisk-cache-stats" 
[cache_sr] in
+               let assoc_list = 
+                       List.filter_map (fun line -> try let hd::tl = 
String.split '=' line in Some (hd,String.concat "=" tl) with _ -> None) 
+                               (String.split '\n' cache_stats_out) 
+               in
+               (*debug "assoc_list: [%s]" (String.concat ";" (List.map (fun 
(a,b) -> Printf.sprintf "%s=%s" a b) assoc_list));*)
+               { time = timestamp;
+                 cache_size_raw = Int64.of_string (List.assoc 
"TOTAL_CACHE_UTILISATION" assoc_list);
+                 cache_hits_raw = Int64.of_string (List.assoc 
"TOTAL_CACHE_HITS" assoc_list);
+                 cache_misses_raw = Int64.of_string (List.assoc 
"TOTAL_CACHE_MISSES" assoc_list); }
+       in
+
+       let get_dss cache_sr oldvals newvals =
+               [ 
+                       (Host,ds_make ~name:(Printf.sprintf "sr_%s_cache_size" 
cache_sr) ~description:"Size in bytes of the cache SR"
+                               ~value:(Rrd.VT_Int64 newvals.cache_size_raw) 
~ty:Rrd.Gauge ~min:0.0 ~default:true ());
+                       (Host,ds_make ~name:(Printf.sprintf "sr_%s_cache_hits" 
cache_sr) ~description:"Hits per second of the cache"
+                               ~value:(Rrd.VT_Int64 (Int64.div (Int64.sub 
newvals.cache_hits_raw oldvals.cache_hits_raw)
+                                       (Int64.of_float (newvals.time -. 
oldvals.time))))
+                               ~ty:Rrd.Gauge ~min:0.0 ~default:true ());
+                       (Host,ds_make ~name:(Printf.sprintf 
"sr_%s_cache_misses" cache_sr) ~description:"Misses per second of the cache"
+                               ~value:(Rrd.VT_Int64 (Int64.div (Int64.sub 
newvals.cache_misses_raw oldvals.cache_misses_raw) 
+                                       (Int64.of_float (newvals.time -. 
oldvals.time))))
+                               ~ty:Rrd.Gauge ~min:0.0 ~default:true ()) ]
+       in
+
+       match !last_cache_stats,cache_sr_opt with 
+               | None, None -> 
+                       []
+               | None, Some cache_sr ->
+                       let stats = do_read cache_sr in
+                       last_cache_stats := Some stats;
+                       []
+               | Some oldstats, None ->
+                       last_cache_stats := None;
+                       []
+               | Some oldstats, Some cache_sr ->
+                       if timestamp -. oldstats.time > 55.0 then begin
+                               let newstats = do_read cache_sr in
+                               last_cache_stats := Some newstats;
+                               let dss = get_dss cache_sr oldstats newstats in
+                               cached_cache_dss := dss;
+                               dss
+                       end else !cached_cache_dss
+
       
 let read_all_dom0_stats __context =
   let handle_exn log f default =
@@ -452,6 +523,7 @@
        handle_exn "read_mem_metrics" (fun ()->read_mem_metrics xc) [];
         vcpus;
        vifs;
+       handle_exn "cache_stats" (fun () -> read_cache_stats timestamp) [];
        handle_exn "update_pcpus" (fun ()->update_pcpus xc) [];
        handle_exn "update_vbds" (fun ()->update_vbds domains) [];
        handle_exn "update_loadavg" (fun ()-> [ update_loadavg () ]) [];
diff -r 243ef81d48f7 -r c9096a7ecb48 ocaml/xapi/xapi_host.ml
--- a/ocaml/xapi/xapi_host.ml
+++ b/ocaml/xapi/xapi_host.ml
@@ -1376,7 +1376,8 @@
                let pbd_host = Db.PBD.get_host ~__context ~self:(List.hd pbds) 
in
                if pbd_host <> host then raise (Api_errors.Server_error 
(Api_errors.host_cannot_see_SR,[Ref.string_of host; Ref.string_of sr]));
                Db.Host.set_local_cache_sr ~__context ~self:host ~value:sr;
-               Db.SR.set_local_cache_enabled ~__context ~self:sr ~value:true
+               Db.SR.set_local_cache_enabled ~__context ~self:sr ~value:true;
+               Monitor.set_cache_sr (Db.SR.get_uuid ~__context ~self:sr);      
        
        end else begin
                raise (Api_errors.Server_error 
(Api_errors.sr_operation_not_supported,[]))
        end
@@ -1385,6 +1386,7 @@
        assert_bacon_mode ~__context ~host;
        let sr = Db.Host.get_local_cache_sr ~__context ~self:host in
        Db.Host.set_local_cache_sr ~__context ~self:host ~value:Ref.null;
+       Monitor.unset_cache_sr ();
        try Db.SR.set_local_cache_enabled ~__context ~self:sr ~value:false with 
_ -> ()

Attachment: xen-api2.hg-2.patch
Description: Text Data

_______________________________________________
xen-api mailing list
xen-api@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/mailman/listinfo/xen-api