diff options
Diffstat (limited to 'lib/os_mon')
| -rw-r--r-- | lib/os_mon/c_src/memsup.c | 2 | ||||
| -rw-r--r-- | lib/os_mon/doc/src/notes.xml | 28 | ||||
| -rw-r--r-- | lib/os_mon/src/disksup.erl | 6 | ||||
| -rw-r--r-- | lib/os_mon/test/disksup_SUITE.erl | 295 | ||||
| -rw-r--r-- | lib/os_mon/vsn.mk | 2 |
5 files changed, 171 insertions, 162 deletions
diff --git a/lib/os_mon/c_src/memsup.c b/lib/os_mon/c_src/memsup.c index 8e610f951b..b5114d10ed 100644 --- a/lib/os_mon/c_src/memsup.c +++ b/lib/os_mon/c_src/memsup.c @@ -1,7 +1,7 @@ /* * %CopyrightBegin% * - * Copyright Ericsson AB 1996-2012. All Rights Reserved. + * Copyright Ericsson AB 1996-2013. All Rights Reserved. * * The contents of this file are subject to the Erlang Public License, * Version 1.1, (the "License"); you may not use this file except in diff --git a/lib/os_mon/doc/src/notes.xml b/lib/os_mon/doc/src/notes.xml index 61eac937e1..2206f93d34 100644 --- a/lib/os_mon/doc/src/notes.xml +++ b/lib/os_mon/doc/src/notes.xml @@ -30,6 +30,34 @@ </header> <p>This document describes the changes made to the OS_Mon application.</p> +<section><title>Os_Mon 2.2.12</title> + + <section><title>Fixed Bugs and Malfunctions</title> + <list> + <item> + <p> + Compilation fixes for NetBSD. Thanks to YAMAMOTO Takashi.</p> + <p> + Own Id: OTP-10941</p> + </item> + </list> + </section> + + + <section><title>Improvements and New Features</title> + <list> + <item> + <p> + Fixed disksup:get_disk_data for SUSv3, specifically OS X + ML. Thanks to Sriram Melkote.</p> + <p> + Own Id: OTP-10945</p> + </item> + </list> + </section> + +</section> + <section><title>Os_Mon 2.2.11</title> <section><title>Fixed Bugs and Malfunctions</title> diff --git a/lib/os_mon/src/disksup.erl b/lib/os_mon/src/disksup.erl index f906f33d32..278da26a20 100644 --- a/lib/os_mon/src/disksup.erl +++ b/lib/os_mon/src/disksup.erl @@ -1,7 +1,7 @@ %% %% %CopyrightBegin% %% -%% Copyright Ericsson AB 1996-2012. All Rights Reserved. +%% Copyright Ericsson AB 1996-2013. All Rights Reserved. %% %% The contents of this file are subject to the Erlang Public License, %% Version 1.1, (the "License"); you may not use this file except in @@ -263,10 +263,10 @@ check_disk_space({unix, dragonfly}, Port, Threshold) -> Result = my_cmd("/bin/df -k -t ufs,hammer", Port), check_disks_solaris(skip_to_eol(Result), Threshold); check_disk_space({unix, freebsd}, Port, Threshold) -> - Result = my_cmd("/bin/df -k -t ufs", Port), + Result = my_cmd("/bin/df -k -l", Port), check_disks_solaris(skip_to_eol(Result), Threshold); check_disk_space({unix, openbsd}, Port, Threshold) -> - Result = my_cmd("/bin/df -k -t ffs", Port), + Result = my_cmd("/bin/df -k -l", Port), check_disks_solaris(skip_to_eol(Result), Threshold); check_disk_space({unix, netbsd}, Port, Threshold) -> Result = my_cmd("/bin/df -k -t ffs", Port), diff --git a/lib/os_mon/test/disksup_SUITE.erl b/lib/os_mon/test/disksup_SUITE.erl index c1ff2c6afc..9c65d8b692 100644 --- a/lib/os_mon/test/disksup_SUITE.erl +++ b/lib/os_mon/test/disksup_SUITE.erl @@ -34,11 +34,11 @@ -define(default_timeout, ?t:minutes(1)). init_per_suite(Config) when is_list(Config) -> - ?line ok = application:start(os_mon), + ok = application:start(os_mon), Config. end_per_suite(Config) when is_list(Config) -> - ?line ok = application:stop(os_mon), + ok = application:stop(os_mon), Config. init_per_testcase(unavailable, Config) -> @@ -78,80 +78,71 @@ end_per_group(_GroupName, Config) -> Config. -api(suite) -> - []; -api(doc) -> - ["Test of API functions"]; +api(suite) -> []; +api(doc) -> ["Test of API functions"]; api(Config) when is_list(Config) -> %% get_disk_data() - ?line [{Id, KByte, Capacity}|_] = disksup:get_disk_data(), - ?line true = io_lib:printable_list(Id), - ?line true = is_integer(KByte), - ?line true = is_integer(Capacity), - ?line true = KByte>0, - ?line true = Capacity>0, + [{Id,KByte,Capacity}|_] = get_disk_data(), + true = io_lib:printable_list(Id), + true = is_integer(KByte), + true = is_integer(Capacity), + true = Capacity>0, + true = KByte>0, %% get_check_interval() - ?line 1800000 = disksup:get_check_interval(), + 1800000 = disksup:get_check_interval(), %% set_check_interval(Minutes) - ?line ok = disksup:set_check_interval(20), - ?line 1200000 = disksup:get_check_interval(), - ?line {'EXIT',{badarg,_}} = (catch disksup:set_check_interval(0.5)), - ?line 1200000 = disksup:get_check_interval(), - ?line ok = disksup:set_check_interval(30), + ok = disksup:set_check_interval(20), + 1200000 = disksup:get_check_interval(), + {'EXIT',{badarg,_}} = (catch disksup:set_check_interval(0.5)), + 1200000 = disksup:get_check_interval(), + ok = disksup:set_check_interval(30), %% get_almost_full_threshold() - ?line 80 = disksup:get_almost_full_threshold(), + 80 = disksup:get_almost_full_threshold(), %% set_almost_full_threshold(Float) - ?line ok = disksup:set_almost_full_threshold(0.90), - ?line 90 = disksup:get_almost_full_threshold(), - ?line {'EXIT',{badarg,_}} = + ok = disksup:set_almost_full_threshold(0.90), + 90 = disksup:get_almost_full_threshold(), + {'EXIT',{badarg,_}} = (catch disksup:set_almost_full_threshold(-0.5)), - ?line 90 = disksup:get_almost_full_threshold(), - ?line ok = disksup:set_almost_full_threshold(0.80), + 90 = disksup:get_almost_full_threshold(), + ok = disksup:set_almost_full_threshold(0.80), ok. -config(suite) -> - []; -config(doc) -> - ["Test configuration"]; +config(suite) -> []; +config(doc) -> ["Test configuration"]; config(Config) when is_list(Config) -> %% Change configuration parameters and make sure change is reflected %% when disksup is restarted - ?line ok = - application:set_env(os_mon, disk_space_check_interval, 29), - ?line ok = - application:set_env(os_mon, disk_almost_full_threshold, 0.81), + ok = application:set_env(os_mon, disk_space_check_interval, 29), + ok = application:set_env(os_mon, disk_almost_full_threshold, 0.81), - ?line ok = supervisor:terminate_child(os_mon_sup, disksup), - ?line {ok, _Child1} = supervisor:restart_child(os_mon_sup, disksup), + ok = supervisor:terminate_child(os_mon_sup, disksup), + {ok, _Child1} = supervisor:restart_child(os_mon_sup, disksup), - ?line 1740000 = disksup:get_check_interval(), - ?line 81 = disksup:get_almost_full_threshold(), + 1740000 = disksup:get_check_interval(), + 81 = disksup:get_almost_full_threshold(), %% Also try this with bad parameter values, should be ignored - ?line ok = + ok = application:set_env(os_mon, disk_space_check_interval, 0.5), - ?line ok = + ok = application:set_env(os_mon, disk_almost_full_threshold, -0.81), - ?line ok = supervisor:terminate_child(os_mon_sup, disksup), - ?line {ok, _Child2} = supervisor:restart_child(os_mon_sup, disksup), + ok = supervisor:terminate_child(os_mon_sup, disksup), + {ok, _Child2} = supervisor:restart_child(os_mon_sup, disksup), - ?line 1800000 = disksup:get_check_interval(), - ?line 80 = disksup:get_almost_full_threshold(), + 1800000 = disksup:get_check_interval(), + 80 = disksup:get_almost_full_threshold(), %% Reset configuration parameters - ?line ok = - application:set_env(os_mon, disk_space_check_interval, 30), - ?line ok = - application:set_env(os_mon, disk_almost_full_threshold, 0.80), - + ok = application:set_env(os_mon, disk_space_check_interval, 30), + ok = application:set_env(os_mon, disk_almost_full_threshold, 0.80), ok. %%---------------------------------------------------------------------- @@ -159,24 +150,22 @@ config(Config) when is_list(Config) -> %% changes too much during its course, or if there are timing problems %% with the alarm_handler receiving the alarms too late %%---------------------------------------------------------------------- -alarm(suite) -> - []; -alarm(doc) -> - ["Test that alarms are set and cleared"]; +alarm(suite) -> []; +alarm(doc) -> ["Test that alarms are set and cleared"]; alarm(Config) when is_list(Config) -> %% Find out how many disks exceed the threshold %% and make sure the corresponding number of alarms is set - ?line Threshold1 = disksup:get_almost_full_threshold(), % 80 - ?line Data1 = disksup:get_disk_data(), - ?line Over1 = over_threshold(Data1, Threshold1), - ?line Alarms1 = get_alarms(), + Threshold1 = disksup:get_almost_full_threshold(), % 80 + Data1 = disksup:get_disk_data(), + Over1 = over_threshold(Data1, Threshold1), + Alarms1 = get_alarms(), if Over1==length(Alarms1) -> - ?line true; + true; true -> dump_info(), - ?line ?t:fail({bad_alarms, Threshold1, Data1, Alarms1}) + ?t:fail({bad_alarms, Threshold1, Data1, Alarms1}) end, %% Try to find a disk with space usage below Threshold1, @@ -187,24 +176,24 @@ alarm(Config) when is_list(Config) -> true -> false end end, - ?line case until(Fun1, Data1) of + case until(Fun1, Data1) of {_, _, Cap1} -> Threshold2 = Cap1-1, - ?line ok = + ok = disksup:set_almost_full_threshold(Threshold2/100), - ?line disksup ! timeout, % force a disk check - ?line Data2 = disksup:get_disk_data(), - ?line Over2 = over_threshold(Data2, Threshold2), - ?line Alarms2 = get_alarms(), + disksup ! timeout, % force a disk check + Data2 = disksup:get_disk_data(), + Over2 = over_threshold(Data2, Threshold2), + Alarms2 = get_alarms(), if Over2==length(Alarms2), Over2>Over1 -> - ?line true; + true; true -> dump_info(), - ?line ?t:fail({bad_alarms, Threshold2, Data2, Alarms2}) + ?t:fail({bad_alarms, Threshold2, Data2, Alarms2}) end; false -> - ?line ignore + ignore end, %% Find out the highest space usage among all disks @@ -216,40 +205,35 @@ alarm(Config) when is_list(Config) -> true -> MaxAcc end end, - ?line case lists:foldl(Fun2, 0, Data1) of + case lists:foldl(Fun2, 0, Data1) of Max when Max<100 -> Threshold3 = Max+1, - ?line ok = - disksup:set_almost_full_threshold(Threshold3/100), - ?line disksup ! timeout, % force a disk check - ?line Data3 = disksup:get_disk_data(), - ?line Over3 = over_threshold(Data3, Threshold3), - ?line Alarms3 = get_alarms(), + ok = disksup:set_almost_full_threshold(Threshold3/100), + disksup ! timeout, % force a disk check + Data3 = disksup:get_disk_data(), + Over3 = over_threshold(Data3, Threshold3), + Alarms3 = get_alarms(), if Over3==0, length(Alarms3)==0 -> - ?line ok; + ok; true -> dump_info(), - ?line ?t:fail({bad_alarms, Threshold3, Data3, Alarms3}) + ?t:fail({bad_alarms, Threshold3, Data3, Alarms3}) end; 100 -> - ?line ignore + ignore end, %% Reset threshold - ?line ok = disksup:set_almost_full_threshold(Threshold1/100), - + ok = disksup:set_almost_full_threshold(Threshold1/100), ok. over_threshold(Data, Threshold) -> Data2 = remove_duplicated_disks(lists:keysort(1, Data)), - lists:foldl(fun({_Id, _Kbyte, Cap}, N) when Cap>=Threshold -> - N+1; - (_DiskData, N) -> - N - end, - 0, - Data2). + lists:foldl(fun + ({_Id, _Kbyte, Cap}, N) when Cap>=Threshold -> N+1; + (_DiskData, N) -> N + end, 0, Data2). %% On some platforms (for example MontaVista) data for one disk can be %% "duplicated": @@ -278,33 +262,30 @@ remove_duplicated_disks([]) -> []. get_alarms() -> - lists:filter(fun({{disk_almost_full, _Disk},_}) -> true; - (_) -> false - end, - alarm_handler:get_alarms()). + lists:filter(fun + ({{disk_almost_full, _Disk},_}) -> true; + (_) -> false + end, alarm_handler:get_alarms()). until(Fun, [H|T]) -> case Fun(H) of true -> H; - false -> - until(Fun, T) + false -> until(Fun, T) end; -until(_Fun, []) -> - false. +until(_Fun, []) -> false. -port(suite) -> - []; +port(suite) -> []; port(doc) -> ["Test that disksup handles a terminating port program"]; port(Config) when is_list(Config) -> - ?line Str = os:cmd("ps -ef | grep '[d]isksup'"), + Str = os:cmd("ps -ef | grep '[d]isksup'"), case io_lib:fread("~s ~s", Str) of - {ok, [_Uid,Pid], _Rest} -> + {ok, [_Uid,Pid], _Rest} -> %% Monitor disksup - ?line MonRef = erlang:monitor(process, disksup), - ?line [{_Disk1,Kbyte1,_Cap1}|_] = disksup:get_disk_data(), - ?line true = Kbyte1>0, + MonRef = erlang:monitor(process, disksup), + [{_Disk1,Kbyte1,_Cap1}|_] = disksup:get_disk_data(), + true = Kbyte1>0, %% Kill the port program case os:cmd("kill -9 " ++ Pid) of @@ -315,17 +296,16 @@ port(Config) when is_list(Config) -> {'DOWN', MonRef, _, _, {port_died, _Reason}} -> ok; {'DOWN', MonRef, _, _, Reason} -> - ?line ?t:fail({unexpected_exit_reason, Reason}) + ?t:fail({unexpected_exit_reason, Reason}) after 3000 -> - ?line ?t:fail({still_alive, Str}) + ?t:fail({still_alive, Str}) end, %% Give os_mon_sup time to restart disksup ?t:sleep(?t:seconds(3)), - ?line [{_Disk2,Kbyte2,_Cap2}|_] = - disksup:get_disk_data(), - ?line true = Kbyte2>0, + [{_Disk2,Kbyte2,_Cap2}|_] = disksup:get_disk_data(), + true = Kbyte2>0, ok; @@ -337,68 +317,61 @@ port(Config) when is_list(Config) -> {skip, {os_pid_not_found, Str}} end. -terminate(suite) -> - []; +terminate(suite) -> []; terminate(Config) when is_list(Config) -> - ?line ok = application:set_env(os_mon, start_disksup, false), - ?line ok = supervisor:terminate_child(os_mon_sup, disksup), + ok = application:set_env(os_mon, start_disksup, false), + ok = supervisor:terminate_child(os_mon_sup, disksup), ok. -unavailable(suite) -> - []; +unavailable(suite) -> []; unavailable(doc) -> ["Test correct behaviour when service is unavailable"]; unavailable(Config) when is_list(Config) -> %% Make sure all API functions return their dummy values - ?line [{"none",0,0}] = disksup:get_disk_data(), - ?line 1800000 = disksup:get_check_interval(), - ?line ok = disksup:set_check_interval(5), - ?line 80 = disksup:get_almost_full_threshold(), - ?line ok = disksup:set_almost_full_threshold(0.9), - + [{"none",0,0}] = disksup:get_disk_data(), + 1800000 = disksup:get_check_interval(), + ok = disksup:set_check_interval(5), + 80 = disksup:get_almost_full_threshold(), + ok = disksup:set_almost_full_threshold(0.9), ok. restart(suite) -> []; restart(Config) when is_list(Config) -> - ?line ok = application:set_env(os_mon, start_disksup, true), - ?line {ok, _Pid} = supervisor:restart_child(os_mon_sup, disksup), + ok = application:set_env(os_mon, start_disksup, true), + {ok, _Pid} = supervisor:restart_child(os_mon_sup, disksup), ok. -otp_5910(suite) -> - []; +otp_5910(suite) -> []; otp_5910(doc) -> ["Test that alarms are cleared if disksup crashes or " "if OS_Mon is stopped"]; otp_5910(Config) when is_list(Config) -> %% Make sure disksup sets at least one alarm - ?line Data = disksup:get_disk_data(), - ?line Threshold0 = disksup:get_almost_full_threshold(), - ?line Threshold = case over_threshold(Data, Threshold0) of - 0 -> - [{_Id,_Kbyte,Cap}|_] = Data, - ?line ok = disksup:set_almost_full_threshold((Cap-1)/100), - Cap-1; - _N -> - Threshold0 - end, - ?line ok = application:set_env(os_mon, - disk_almost_full_threshold, - Threshold/100), - ?line disksup ! timeout, % force a disk check - ?line Data2 = disksup:get_disk_data(), - ?line Over = over_threshold(Data2, Threshold), - ?line Alarms = get_alarms(), + Data = disksup:get_disk_data(), + Threshold0 = disksup:get_almost_full_threshold(), + Threshold = case over_threshold(Data, Threshold0) of + 0 -> + [{_Id,_Kbyte,Cap}|_] = Data, + ok = disksup:set_almost_full_threshold((Cap-1)/100), + Cap-1; + _N -> Threshold0 + end, + ok = application:set_env(os_mon, disk_almost_full_threshold, Threshold/100), + disksup ! timeout, % force a disk check + Data2 = disksup:get_disk_data(), + Over = over_threshold(Data2, Threshold), + Alarms = get_alarms(), if Over==0 -> - ?line ?t:fail({threshold_too_low, Data2, Threshold}); + ?t:fail({threshold_too_low, Data2, Threshold}); Over==length(Alarms) -> ok; true -> dump_info(), - ?line ?t:fail({bad_alarms, Threshold, Data2, Alarms}) + ?t:fail({bad_alarms, Threshold, Data2, Alarms}) end, %% Kill disksup @@ -407,34 +380,42 @@ otp_5910(Config) when is_list(Config) -> %% Wait a little to make sure disksup has been restarted, %% then make sure the alarms are set once, but not twice ?t:sleep(?t:seconds(1)), - ?line Data3 = disksup:get_disk_data(), - ?line Alarms2 = get_alarms(), + Data3 = disksup:get_disk_data(), + Alarms2 = get_alarms(), if - length(Alarms2)==length(Alarms) -> - ok; + length(Alarms2)==length(Alarms) -> ok; true -> dump_info(), - ?line ?t:fail({bad_alarms, Threshold, Data3, Alarms,Alarms2}) + ?t:fail({bad_alarms,Threshold,Data3,Alarms,Alarms2}) end, %% Stop OS_Mon and make sure all disksup alarms are cleared - ?line ok = application:stop(os_mon), + ok = application:stop(os_mon), ?t:sleep(?t:seconds(1)), - ?line Alarms3 = get_alarms(), - if - length(Alarms3)==0 -> - ok; - true -> - ?line ?t:fail({alarms_not_cleared, Alarms3}) + Alarms3 = get_alarms(), + case get_alarms() of + [] -> ok; + _ -> ?t:fail({alarms_not_cleared, Alarms3}) end, %% Reset threshold and restart OS_Mon - ?line ok = application:set_env(os_mon, - disksup_almost_full_threshold, 0.8), - ?line ok = disksup:set_almost_full_threshold(0.8), - ?line ok = application:start(os_mon), - + ok = application:set_env(os_mon, disksup_almost_full_threshold, 0.8), + ok = disksup:set_almost_full_threshold(0.8), + ok = application:start(os_mon), ok. dump_info() -> io:format("Status: ~p~n", [sys:get_status(disksup)]). + +% filter get_disk_data and remove entriew with zero capacity +% "non-normal" filesystems report zero capacity +% - Perhaps errorneous 'df -k -l'? +% - Always list filesystems by type '-t ufs,zfs,..' instead? +% It is unclear what the intention was from the beginning. +get_disk_data() -> + get_disk_data(disksup:get_disk_data()). + +get_disk_data([{"none",0,0}=E]) -> [E]; +get_disk_data([{_,_,0}|Es]) -> get_disk_data(Es); +get_disk_data([E|Es]) -> [E|get_disk_data(Es)]; +get_disk_data([]) -> []. diff --git a/lib/os_mon/vsn.mk b/lib/os_mon/vsn.mk index 7d6c5484a7..e3acea0258 100644 --- a/lib/os_mon/vsn.mk +++ b/lib/os_mon/vsn.mk @@ -1 +1 @@ -OS_MON_VSN = 2.2.11 +OS_MON_VSN = 2.2.12 |