[checkmk-commits] Further optimization of ps check for large number of processes

Module: check_mk Branch: master Commit: fc46323af3fafd834b2fc8c095152a7cc1cdebc6 URL: http://git.mathias-kettner.de/git/?p=check_mk.git;a=commit;h=fc46323af3fafd… Author: Mathias Kettner &lt;mk(a)mathias-kettner.de&gt; Date: Fri Oct 16 16:40:20 2015 +0200 Further optimization of ps check for large number of processes --- .werks/2664 | 2 +- checks/ps.include | 50 +++++++++++++++++++++++++++++++++++++------------- 2 files changed, 38 insertions(+), 14 deletions(-) diff --git a/.werks/2664 b/.werks/2664 index 6891d01..29cf879 100644 --- a/.werks/2664 +++ b/.werks/2664 @@ -7,4 +7,4 @@ Date: 1445002617 Class: fix In one situation with about 7500 matching processes out of 8000 the check now saves -about 15% of the CPU ressources. +about 20% of the CPU ressources. diff --git a/checks/ps.include b/checks/ps.include index 89d180c..09e7b8c 100644 --- a/checks/ps.include +++ b/checks/ps.include @@ -86,7 +86,7 @@ def inventory_ps_common(invdata, invrules, info): # First entry in line is the node name or None for non-clusters ps = line[1:] l_user = [userspec] - matches = process_matches(ps, pattern, l_user) + matches = process_matches_for_discovery(ps, pattern, l_user) if matches != False: if len(matches) < num_perc_s: raise MKGeneralException("Invalid entry in inventory_processes: service description '%s' contains " @@ -153,17 +153,7 @@ def escape_regex_chars(match): return r -def process_matches(ps, procname, l_user): - # procname is either: - # 1. a string beginning with ~. Then it is interpreted as regular expression - # that must match the *beginning* of the process line. Please check the output of - # check_mk -d HOSTNAME. Note: groups of whitespaces are reduced to one single - # whitespace! - # 2. a string *not* beginning with ~: It must be equal to the first column - # in the process table (i.e. the process name). No regular expressions are - # applied. A simple string compare is done. - - # agent might output username in brackets in the first columns +def process_matches_for_discovery(ps, procname, l_user): userspec = l_user[0] if ps[0].startswith("(") and ps[0].endswith(")") and len(ps) > 1: addinfo = ps[0][1:-1].split(",") @@ -194,6 +184,40 @@ def process_matches(ps, procname, l_user): return False + +# Returns True or False +def process_matches_for_check(process_line, process_pattern, user_pattern): + + # Check process owner. Note: the first field with process information is + # optional and might be missing (old agents). + if process_line[0].startswith("(") and process_line[0].endswith(")") and len(process_line) > 1: + if user_pattern: + addinfo = process_line[0][1:-1].split(",") + user = addinfo[0] + if user_pattern and user_pattern.startswith('~'): + if not regex(user_pattern[1:]).match(user): + return False + + elif user_pattern and user_pattern != user: + return False + + process_line = process_line[1:] + + # Process name not relevant + if not process_pattern: + return True + + # Regex for complete process command line + elif process_pattern.startswith("~"): + reg = regex(process_pattern[1:]) # skip "~" + return reg.match(" ".join(process_line)) + + # Exact match on name of executable + else: + return process_line[0] == process_pattern + + + # produce text or html output intended for the long output field of a check from details about a # process. # the input is expected to be a list (one per process) of lists (one per data field) of key-value @@ -322,7 +346,7 @@ def check_ps_common(item, params, info, cpu_cores = 1, info_name = "processes", for line in info: node_name = line[0] ps = line[1:] - if process_matches(ps, params.get("process"), [params.get("user")]) != False: + if process_matches_for_check(ps, params.get("process"), params.get("user")): process = [] count += 1 if node_name != None: