[checkmk-commits] Check_MK Git: check_mk: Availability: better state grouping and melting

Module: check_mk Branch: master Commit: d154f19d2dd033a5f27579bada956dc7e600acbc URL: http://git.mathias-kettner.de/git/?p=check_mk.git;a=commit;h=d154f19d2dd033… Author: Mathias Kettner &lt;mk(a)mathias-kettner.de&gt; Date: Wed Feb 6 11:56:16 2013 +0100 Availability: better state grouping and melting --- web/htdocs/views.css | 6 +- web/plugins/views/availability.py | 139 +++++++++++++++++++++++++++---------- 2 files changed, 104 insertions(+), 41 deletions(-) diff --git a/web/htdocs/views.css b/web/htdocs/views.css index 596255f..66921da 100644 --- a/web/htdocs/views.css +++ b/web/htdocs/views.css @@ -482,8 +482,8 @@ table.data.availability td.number { } table.data td.flapping, table.timeline td.flapping { - background-image: url("images/bg_flapping.png"); - background-repeat: repeat; + /* background-image: url("images/bg_flapping.png"); + background-repeat: repeat; */ background-color: #f0f; } @@ -499,7 +499,7 @@ table.data td.unmonitored, table.timeline td.unmonitored { color: #888; } table.data.availability tr.odd0 td.number { - background-image: url("images/white_20percent.png"); + background-image: url("images/white_30percent.png"); background-repeat: repeat; } diff --git a/web/plugins/views/availability.py b/web/plugins/views/availability.py index 72c7158..ab71e32 100644 --- a/web/plugins/views/availability.py +++ b/web/plugins/views/availability.py @@ -119,13 +119,58 @@ avoption_entries = [ Checkbox(label = _("Consider notification period")), ), ( "unmonitored", - Checkbox(label = _("Consider unmonitored time")), + Checkbox(label = _("Include unmonitored time")), ), ], optional_keys = False, ), ), + # Optionally group some states togehter + ( "state_grouping", + "double", + Dictionary( + title = _("Status Grouping"), + columns = 2, + elements = [ + ( "warn", + DropdownChoice( + label = _("Treat Warning as: "), + choices = [ + ( "ok", _("OK") ), + ( "warn", _("WARN") ), + ( "crit", _("CRIT") ), + ( "unknown", _("UNKNOWN") ), + ] + ), + ), + ( "unknown", + DropdownChoice( + label = _("Treat Unknown as: "), + choices = [ + ( "ok", _("OK") ), + ( "warn", _("WARN") ), + ( "crit", _("CRIT") ), + ( "unknown", _("UNKNOWN") ), + ] + ), + ), + ( "host_down", + DropdownChoice( + label = _("Treat Host Down as: "), + choices = [ + ( "ok", _("OK") ), + ( "warn", _("WARN") ), + ( "crit", _("CRIT") ), + ( "unknown", _("UNKNOWN") ), + ( "host_down", _("Host Down") ), + ] + ), + ), + ], + optional_keys = False, + ), + ), # Format of numbers ( "timeformat", "single", @@ -145,21 +190,14 @@ avoption_entries = [ ), - # Optionally group some states togehter - ( "state_grouping", + # Short time intervals + ( "short_intervals", "single", - Dictionary( - title = _("Status Grouping"), - columns = 2, - elements = [ - ( "warn_is_ok", - Checkbox(label = _("Treat WARN as OK")), - ), - ( "unknown_is_crit", - Checkbox(label = _("Treat UNKNOWN as CRIT")), - ), - ], - optional_keys = False, + Integer( + title = _("Short Time Intervalls"), + minvalue = 0, + unit = _("sec"), + label = _("Ignore intervals shorter or equal"), ), ), ] @@ -178,9 +216,11 @@ def render_availability_options(): "timeformat" : "percentage_2", "rangespec" : "d0", "state_grouping" : { - "warn_is_ok" : False, - "unknown_is_crit" : False, + "warn" : "warn", + "unknown" : "unknown", + "host_down" : "host_down", }, + "short_intervals" : 0, }) is_open = False @@ -380,12 +420,12 @@ def do_render_availability(datasource, filterheaders, avoptions, only_sites, lim # 2.2.2.2.3 "crit" # 2.2.2.2.4 "unknown" availability = [] - timeline_rows = [] - timeline_considered_duration = 0 # Note: in case of timeline, we have data from exacly one host/service for site_host, site_host_entry in by_host.iteritems(): for service, service_entry in site_host_entry.iteritems(): - states = {} + + # First compute timeline + timeline_rows = [] considered_duration = 0 for span in service_entry: state = span["state"] @@ -406,20 +446,32 @@ def do_render_availability(datasource, filterheaders, avoptions, only_sites, lim s = { 0: "ok", 1:"warn", 2:"crit", 3:"unknown" }[state] else: s = { 0: "up", 1:"down", 2:"unreach"}[state] - if avoptions["state_grouping"]["warn_is_ok"] and s == "warn": - s = "ok" - elif avoptions["state_grouping"]["unknown_is_crit"] and s == "unknown": - s = "crit" - # TODO: host_down as crit + if s == "warn": + s = avoptions["state_grouping"]["warn"] + elif s == "unknown": + s = avoptions["state_grouping"]["unknown"] + elif s == "host_down": + s = avoptions["state_grouping"]["host_down"] + + considered_duration += span["duration"] + timeline_rows.append((span, s)) + + # Now merge consecutive rows with identical state + merge_timeline(timeline_rows) + # Melt down short intervals + if avoptions["short_intervals"]: + melt_short_intervals(timeline_rows, avoptions["short_intervals"]) + + # Condense into availability + states = {} + for span, s in timeline_rows: states.setdefault(s, 0) states[s] += span["duration"] - considered_duration += span["duration"] - if timeline: - timeline_rows.append((span, s)) - timeline_considered_duration = considered_duration + availability.append([site_host[0], site_host[1], service, states, considered_duration]) + # Prepare number format function from_time, until_time = range duration = until_time - from_time @@ -446,7 +498,7 @@ def do_render_availability(datasource, filterheaders, avoptions, only_sites, lim return "%02d:%02d:%02d" % (hours, minn, sec) if timeline: - render_timeline(timeline_rows, from_time, until_time, timeline_considered_duration, timeline, range_title, render_number) + render_timeline(timeline_rows, from_time, until_time, considered_duration, timeline, range_title, render_number) else: render_availability_table(availability, from_time, until_time, range_title, has_service, avoptions, render_number) @@ -455,9 +507,6 @@ def render_timeline(timeline_rows, from_time, until_time, considered_duration, t html.write('<div class=info>%s</div>' % _("No information available")) return - # More rows with identical state - merge_timeline(timeline_rows) - # Timeformat: show date only if the displayed time range spans over # more than one day. format = "%H:%M:%S" @@ -526,8 +575,26 @@ def merge_timeline(entries): else: n += 1 +def melt_short_intervals(entries, duration): + n = 1 + need_merge = False + while n < len(entries) - 1: + if entries[n][0]["duration"] <= duration and \ + entries[n-1][1] == entries[n+1][1]: + entries[n] = (entries[n][0], entries[n-1][1]) + need_merge = True + n += 1 + + # Due to melting, we need to merge again + if need_merge: + merge_timeline(entries) + melt_short_intervals(entries, duration) def render_availability_table(availability, from_time, until_time, range_title, has_service, avoptions, render_number): + # Some columns might be unneeded due to state treatment options + sg = avoptions["state_grouping"] + sgs = [ sg["warn"], sg["unknown"], sg["host_down"] ] + # Render the stuff availability.sort() table.begin(_("Availability") + " " + range_title, css="availability") @@ -572,13 +639,9 @@ def render_availability_table(availability, from_time, until_time, range_title, continue elif sid == "in_downtime" and not avoptions["consider"]["downtime"]: continue - elif sid == "host_down" and not avoptions["consider"]["host_down"]: - continue elif sid == "unmonitored" and not avoptions["consider"]["unmonitored"]: continue - elif sid == "warn" and avoptions["state_grouping"]["warn_is_ok"]: - continue - elif sid == "unknown" and avoptions["state_grouping"]["unknown_is_crit"]: + elif sid in [ "warn", "unknown", "host_down" ] and sid not in sgs: continue number = states.get(sid, 0) if not number: