mirror of https://github.com/schoebel/mars
monitoring: updates
This commit is contained in:
Joerg Mann
Thomas Schoebel-Theuer
parent
4c36e803f3
commit
cdf8d61499
|
|
@ -1,6 +1,5 @@
|
|||
#!/usr/bin/perl -w
|
||||
# (c) 2012/2013 Joerg Mann / 1&1 Internet AG
|
||||
# released under GPL
|
||||
#
|
||||
# last update at now ...
|
||||
|
||||
|
|
@ -21,7 +20,7 @@ use File::Basename;
|
|||
binmode STDOUT, ":utf8";
|
||||
|
||||
### defaults
|
||||
my $version = "0.072m";
|
||||
my $version = "0.072q";
|
||||
my $alife_timeout = "30"; # sec for remote-nodes timeout
|
||||
my $is_tty = 0;
|
||||
my $mars_dir = '/mars';
|
||||
|
|
@ -275,7 +274,7 @@ sub display_partner {
|
|||
$PUDevice = ( grep { /^$PUDevice / } <$fh> )[0];
|
||||
if ( $PUDevice ) {
|
||||
$PUDevice = ( split / /, $PUDevice )[1];
|
||||
print_screen " and mountet as $PUDevice\n", "$Color_blue";
|
||||
print_screen " and mounted as $PUDevice\n", "$Color_blue";
|
||||
} else {
|
||||
print_screen "\n\t\t---> TODO: enable to mount\n", "$Color_green";
|
||||
}
|
||||
|
|
@ -292,16 +291,18 @@ sub display_partner {
|
|||
### check sync
|
||||
### sync - status
|
||||
my $PSyncsize = check_link "$mars_dir/$PRes/syncstatus-$PName";
|
||||
my $SStatus = sprintf("%.2f", ($PSyncsize / $PSize * 100));
|
||||
my $SStatus = sprintf ("%.2f", ($PSyncsize / $PSize * 100));
|
||||
print_screen (sprintf "\tSync : %s bytes (%.3fTB) synced = ", $PSyncsize, ( $PSyncsize/1024/1024/1024/1024));
|
||||
|
||||
### sync - speed
|
||||
my $SSpeed = check_link "$mars_dir/$PRes/actual-$PName/sync_rate";
|
||||
$SSpeed = sprintf ("%.2f", $SSpeed / 1024 / 1024);
|
||||
my $SEndTime = ($PSize - $PSyncsize ) / 1024 / 1024 / 1024;
|
||||
if ( $SSpeed eq "0.00" ) {
|
||||
$SSpeed = "%";
|
||||
} else {
|
||||
$SSpeed = "%, by $SSpeed mb/s";
|
||||
$SEndTime = sprintf ("%.2f", $SEndTime / $SSpeed / 60);
|
||||
$SSpeed = "%, by $SSpeed gb/s (hypothetically ends in $SEndTime min)";
|
||||
}
|
||||
|
||||
### sync - results
|
||||
|
|
@ -340,7 +341,7 @@ sub display_partner {
|
|||
print_screen (sprintf "\tLogfile : %s bytes (%.3fGB) in ", $PLogSize, ( $PLogSize/1024/1024/1024 ));
|
||||
print_screen "$PLogName", "$Color_green";
|
||||
print_screen " active";
|
||||
print_screen ", received with $LogSpeed mb/s" if ( $LogSpeed ne "0.00" );
|
||||
print_screen ", received with $LogSpeed gb/s" if ( $LogSpeed ne "0.00" );
|
||||
print_screen "\n";
|
||||
if ( $Ljoined eq "0" || $PLogSize eq "0.0001" ) {
|
||||
print_screen "\t\t---> WORK: Logfile wait for starting ...\n", "$Color_red";
|
||||
|
|
@ -365,7 +366,7 @@ sub display_partner {
|
|||
if ( $RSpeed eq "0.00" ) {
|
||||
$RSpeed = "%";
|
||||
} else {
|
||||
$RSpeed = "%, by $RSpeed mb/s";
|
||||
$RSpeed = "%, by $RSpeed gb/s";
|
||||
}
|
||||
|
||||
### replay - results
|
||||
|
|
@ -426,7 +427,7 @@ sub display_partner {
|
|||
### check switches
|
||||
my $SWStatus;
|
||||
print_screen "\tSwitches: Attach=";
|
||||
if ( readlink "$mars_dir/$PRes/todo-$PName/attach" eq 1 ) {
|
||||
if ( readlink "$mars_dir/$PRes/todo-$PName/attach" eq 1 ) { ### Use of uninitialized value in string
|
||||
print_screen "on", "$Color_green";
|
||||
monitoring "", "attached";
|
||||
} else {
|
||||
|
|
@ -435,7 +436,7 @@ sub display_partner {
|
|||
}
|
||||
print_screen " [masked:" if ( $ActStatus eq 1 );
|
||||
print_screen " Connect=";
|
||||
if ( readlink "$mars_dir/$PRes/todo-$PName/connect" eq 1 ) {
|
||||
if ( readlink "$mars_dir/$PRes/todo-$PName/connect" eq 1 ) { ### Use of uninitialized value in string
|
||||
print_screen "on", "$Color_green";
|
||||
monitoring "", "connected";
|
||||
} else {
|
||||
|
|
@ -443,7 +444,7 @@ sub display_partner {
|
|||
monitoring "SwitchOff", "connect off";
|
||||
}
|
||||
print_screen " Sync=";
|
||||
if ( readlink "$mars_dir/$PRes/todo-$PName/sync" eq 1 ) {
|
||||
if ( readlink "$mars_dir/$PRes/todo-$PName/sync" eq 1 ) { ### Use of uninitialized value in string
|
||||
print_screen "on", "$Color_green";
|
||||
monitoring "", "synced";
|
||||
} else {
|
||||
|
|
@ -451,7 +452,7 @@ sub display_partner {
|
|||
monitoring "SwitchOff", "sync off";
|
||||
}
|
||||
print_screen " AllowReplay=" ;
|
||||
if ( readlink "$mars_dir/$PRes/todo-$PName/allow-replay" eq 1 ) {
|
||||
if ( readlink "$mars_dir/$PRes/todo-$PName/allow-replay" eq 1 ) { ### Use of uninitialized value in string
|
||||
print_screen "on", "$Color_green";
|
||||
monitoring "", "replayed";
|
||||
} else {
|
||||
|
|
@ -971,42 +972,45 @@ sub check_systemstatus {
|
|||
print_screen "---> Systemdata <---\n", "$Color_blue bold";
|
||||
|
||||
### text / sol-file / sol-einheit / ist-file / ist-einheit
|
||||
check_limit "-> AVG Limit", "loadavg_limit", "loadavg";
|
||||
check_limit "-> AVG Limit", "loadavg_limit", "loadavg";
|
||||
print "\n";
|
||||
check_limit "-> Memory Limit", "mem_limit_percent", "%", "mem_used_raw_kb", "kb";
|
||||
check_limit "-> Memory Limit", "mem_limit_percent", "%", "mem_used_raw_kb", "kb";
|
||||
print "\n";
|
||||
check_limit "-> Traffic Limit", "tuning/traffic_limit_kb", "kb/s", "tuning/traffic_rate_kb", "kb/s";
|
||||
check_limit "-> Traffic Limit Client", "traffic_tuning/client_role_traffic_ratelimit_kb", "kb/s", "traffic_tuning/client_role_traffic_rate_kb", "kb/s";
|
||||
print "\n";
|
||||
check_limit "-> Writeback Limit", "tuning/writeback_limit_kb", "kb/s", "tuning/writeback_rate_kb", "kb/s";
|
||||
check_limit "", "", "", "tuning/writeback_until_percent", "%";
|
||||
check_limit "-> Traffic Limit Server", "traffic_tuning/server_role_traffic_ratelimit_kb", "kb/s", "traffic_tuning/server_role_traffic_rate_kb", "kb/s";
|
||||
print "\n";
|
||||
check_limit "-> Server-IO Limit", "tuning/server_io_limit_kb", "kb/s", "tuning/server_io_rate_kb", "kb/s";
|
||||
check_limit " Flying IO", "", "", "io_flying_count";
|
||||
check_limit "-> Traffic Writeback", "traffic_tuning/writeback_ratelimit_kb", "kb/s", "traffic_tuning/writeback_rate_kb", "kb/s";
|
||||
print "\n";
|
||||
check_limit "-> Copy Read: Prio", "copy_read_prio", "";
|
||||
check_limit " Flying IO", "", "", "copy_read_max_fly", "";
|
||||
check_limit "-> I/O Writeback", "io_tuning/writeback_ratelimit_kb", "kb/s", "io_tuning/writeback_rate_kb", "kb/s";
|
||||
print "\n";
|
||||
check_limit "-> Copy Write: Prio", "copy_write_prio", "";
|
||||
check_limit " Flying IO", "", "", "copy_write_max_fly", "";
|
||||
check_limit "-> Server-IO Limit", "tuning/server_io_limit_kb", "kb/s", "tuning/server_io_rate_kb", "kb/s";
|
||||
check_limit " Flying IO", "", "", "io_flying_count";
|
||||
print "\n";
|
||||
check_limit "-> LoggerMemory", "", "", "logger_mem_used_kb", "kb";
|
||||
check_limit "-> Copy Read: Prio", "copy_read_prio", "";
|
||||
check_limit " Flying IO", "", "", "copy_read_max_fly";
|
||||
print "\n";
|
||||
check_limit "-> Copy Write: Prio", "copy_write_prio", "";
|
||||
check_limit " Flying IO", "", "", "copy_write_max_fly";
|
||||
print "\n";
|
||||
check_limit "-> FreeSpaceLimit LogRotate", "logrot_auto_gb", "gb";
|
||||
check_limit "-> LoggerMemory", "", "", "logger_mem_used_kb", "kb";
|
||||
print "\n";
|
||||
check_limit "-> Network-IO-Timeout", "network_io_timeout", "sec";
|
||||
check_limit "-> FreeSpaceLimit LogRotate", "logrot_auto_gb", "gb";
|
||||
print "\n";
|
||||
check_limit "-> Clear Page Cache", "", "", "mapfree_period_sec", "sec";
|
||||
check_limit "-> Network-IO-Timeout", "network_io_timeout", "sec";
|
||||
print "\n";
|
||||
check_limit "-> Statusfile Rollover", "", "", "statusfiles_rollover_sec", "sec";
|
||||
check_limit "-> Clear Page Cache", "", "", "mapfree_period_sec", "sec";
|
||||
print "\n";
|
||||
check_limit "-> Modus: Fast Full Sync", "", "", "do_fast_fullsync", "on/off";
|
||||
check_limit " AIO Sync", "", "", "aio_sync_mode", "on/off";
|
||||
check_limit " Delay say Overflow", "", "", "delay_say_on_overflow", "on/off";
|
||||
check_limit " Emergency", "", "", "mars_emergency_mode", "on/off";
|
||||
check_limit " Logger Resume", "", "", "logger_resume", "on/off";
|
||||
check_limit "-> Statusfile Rollover", "", "", "statusfiles_rollover_sec", "sec";
|
||||
print "\n";
|
||||
check_limit "-> LamportClockDifferenz", "lamport_clock", "sec";
|
||||
check_limit " Mars Port", "mars_port", "";
|
||||
check_limit "-> Modus: Fast Full Sync", "", "", "do_fast_fullsync", "on/off";
|
||||
check_limit " AIO Sync", "", "", "aio_sync_mode", "on/off";
|
||||
check_limit " Delay say Overflow", "", "", "delay_say_on_overflow", "on/off";
|
||||
check_limit " Emergency", "", "", "mars_emergency_mode", "on/off";
|
||||
check_limit " Logger Resume", "", "", "logger_resume", "on/off";
|
||||
print "\n";
|
||||
check_limit "-> LamportClockDifferenz", "lamport_clock", "sec";
|
||||
check_limit " Mars Port", "mars_port", "";
|
||||
print "\n";
|
||||
|
||||
my $mars_disk_space = `df '$mars_dir' | grep '$mars_dir'| awk '{print \$2}'`;
|
||||
|
|
@ -1071,13 +1075,14 @@ while(1) {
|
|||
if ( $params->{'system'} ) {
|
||||
check_systemstatus;
|
||||
|
||||
### check system params
|
||||
check_disk_is_full;
|
||||
check_jammed;
|
||||
}
|
||||
|
||||
|
||||
##########################################################################
|
||||
### check system params
|
||||
check_disk_is_full;
|
||||
check_jammed;
|
||||
|
||||
### check resources
|
||||
print_screen "---> Resources <---\n", "$Color_blue bold";
|
||||
check_ressource;
|
||||
|
|
|
|||
|
|
@ -1,91 +1,28 @@
|
|||
#!/bin/bash
|
||||
#
|
||||
# Copyright 2013 Joerg Mann / 1&1 Internet AG
|
||||
#
|
||||
# Copying and distribution of this file, with or without modification,
|
||||
# are permitted in any medium without royalty provided the copyright
|
||||
# notice and this notice are preserved. This file is offered as-is,
|
||||
# without any warranty.
|
||||
#####################################################################
|
||||
#
|
||||
# use for mars relase 2013/11/19
|
||||
#
|
||||
# v 0.02
|
||||
# modify by jms at Fri Jan 31 14:11:36 CET 2014
|
||||
# zabbix mars
|
||||
|
||||
LOG="/tmp/zabbix.mars"
|
||||
MDIR="/proc/sys/mars"
|
||||
echo "Date: `date`" >$LOG
|
||||
TEMPFILE="/tmp/zabbix.mars"
|
||||
echo -n >$TEMPFILE
|
||||
|
||||
|
||||
work() {
|
||||
if [ -f $MDIR/$FN ]; then
|
||||
echo "$MDIR/$FN `cat $MDIR/$FN`" >>$LOG
|
||||
else
|
||||
echo "$MDIR/$FN 0" >>$LOG
|
||||
fi
|
||||
}
|
||||
if [ "`lsmod|grep mars|grep -v grep`" == "" ]; then
|
||||
exit 0
|
||||
else
|
||||
# statistic
|
||||
find /proc/sys/mars -name "*" -nowarn -exec echo -n {}" " \; -exec cat {} \; >$TEMPFILE 2>&1
|
||||
|
||||
# status
|
||||
HOSTNAME="`/bin/hostname`"
|
||||
ls -ld /mars/resource-*|awk '{print $9}'|sed -e 's!.*resource-!!g' | \
|
||||
while read RESNAME; do
|
||||
#echo "-> $HOSTNAME - $RESNAME"
|
||||
echo "# actual $RESNAME $HOSTNAME" >>$TEMPFILE
|
||||
ls -ld /mars/resource-$RESNAME/actual-$HOSTNAME/* | sed -e 's!.*/actual-.*/!!g' | awk '{print "status '$RESNAME' " $1 " " $3}' >>$TEMPFILE
|
||||
echo "# todo $RESNAME $HOSTNAME" >>$TEMPFILE
|
||||
ls -ld /mars/resource-$RESNAME/todo-$HOSTNAME/* | sed -e 's!.*/todo-.*/!!g' | awk '{print "status '$RESNAME' " $1 " " $3}' >>$TEMPFILE
|
||||
done
|
||||
|
||||
FN="copy_read_max_fly"; work
|
||||
FN="copy_write_max_fly"; work
|
||||
FN="io_flying_count"; work
|
||||
FN="logger_mem_used_kb"; work
|
||||
FN="logger_replay_timeout_sec"; work
|
||||
FN="mapfree_period_sec"; work
|
||||
FN="mem_used_raw_kb"; work
|
||||
FN="propagate_interval_sec"; work
|
||||
FN="scan_interval_sec"; work
|
||||
FN="statusfiles_rollover_sec"; work
|
||||
FN="sync_flip_interval_sec"; work
|
||||
FN="tuning/aio_io_r_true_hit"; work
|
||||
FN="tuning/aio_io_r_factor_percent"; work
|
||||
FN="tuning/aio_io_w_true_hit"; work
|
||||
FN="tuning/aio_io_w_factor_percent"; work
|
||||
FN="tuning/aio_sync_true_hit"; work
|
||||
FN="tuning/aio_submit_factor_percent"; work
|
||||
FN="tuning/aio_submit_true_hit"; work
|
||||
FN="tuning/aio_sync_factor_percent"; work
|
||||
FN="tuning/bio_io_r_true_hit"; work
|
||||
FN="tuning/bio_io_r_factor_percent"; work
|
||||
FN="tuning/bio_io_w_true_hit"; work
|
||||
FN="tuning/bio_io_w_factor_percent"; work
|
||||
FN="tuning/bio_submit_factor_percent"; work
|
||||
FN="tuning/bio_submit_true_hit"; work
|
||||
FN="tuning/server_io_rate_kb"; work
|
||||
FN="tuning/traffic_rate_kb"; work
|
||||
FN="tuning/writeback_rate_kb"; work
|
||||
FN="tuning/writeback_until_percent"; work
|
||||
FN="fimem_alloc_count"; work
|
||||
FN="mem_alloc_max"; work
|
||||
FN="mem_allow_freelist"; work
|
||||
FN="tuning/write_throttle_count_ops"; work
|
||||
FN="tuning/write_throttle_cumul_kb"; work
|
||||
FN="tuning/write_throttle_end_percent"; work
|
||||
FN="tuning/write_throttle_maxdelay_ms"; work
|
||||
FN="tuning/write_throttle_maxwindow_ms"; work
|
||||
FN="tuning/write_throttle_minwindow_ms"; work
|
||||
FN="tuning/write_throttle_rate_kb"; work
|
||||
FN="tuning/write_throttle_ratelimit_kb"; work
|
||||
FN="tuning/write_throttle_size_threshold_kb"; work
|
||||
FN="tuning/write_throttle_start_percent"; work
|
||||
FN="tuning/writeback_count_ops"; work
|
||||
FN="tuning/writeback_cumul_kb"; work
|
||||
FN="tuning/writeback_maxdelay_ms"; work
|
||||
FN="tuning/writeback_maxwindow_ms"; work
|
||||
FN="tuning/writeback_minwindow_ms"; work
|
||||
FN="tuning/traffic_count_ops"; work
|
||||
FN="tuning/traffic_cumul_kb"; work
|
||||
FN="tuning/traffic_maxdelay_ms"; work
|
||||
FN="tuning/traffic_maxwindow_ms"; work
|
||||
FN="tuning/traffic_minwindow_ms"; work
|
||||
FN="tuning/server_io_count_ops"; work
|
||||
FN="tuning/server_io_cumul_kb"; work
|
||||
FN="tuning/server_io_maxdelay_ms"; work
|
||||
FN="tuning/server_io_maxwindow_ms"; work
|
||||
FN="tuning/server_io_minwindow_ms"; work
|
||||
|
||||
HOSTNAME="`/bin/hostname`"
|
||||
ls -ld /mars/resource-*|awk '{print $9}'|sed -e 's!.*resource-!!g' | \
|
||||
while read RESNAME; do
|
||||
#echo "-> $HOSTNAME - $RESNAME"
|
||||
ls -ld /mars/resource-$RESNAME/actual-$HOSTNAME/* | sed -e 's!.*/actual-.*/!!g' | awk '{print "status '$RESNAME' " $1 " " $3}' >>$LOG
|
||||
ls -ld /mars/resource-$RESNAME/todo-$HOSTNAME/* | sed -e 's!.*/todo-.*/!!g' | awk '{print "status '$RESNAME' " $1 " " $3}' >>$LOG
|
||||
done
|
||||
fi
|
||||
|
|
|
|||
|
|
@ -1,27 +1,6 @@
|
|||
# Copyright 2013 Joerg Mann / 1&1 Internet AG
|
||||
#
|
||||
# Copying and distribution of this file, with or without modification,
|
||||
# are permitted in any medium without royalty provided the copyright
|
||||
# notice and this notice are preserved. This file is offered as-is,
|
||||
# without any warranty.
|
||||
#####################################################################
|
||||
#
|
||||
# zabbix mars config
|
||||
#
|
||||
# mars config
|
||||
# v002
|
||||
|
||||
UserParameter=system.mars[*], echo -n "0"; cat /tmp/zabbix.mars|grep $1|awk '{print $$2}'
|
||||
UserParameter=system.marsstatus[*], echo -n "0"; cat /tmp/zabbix.mars|grep $1|grep ' $2 '|awk '{print $$4}'
|
||||
|
||||
#status data1 device-data1 1
|
||||
#status data1 is-attached 1
|
||||
#status data1 is-copying 0
|
||||
#status data1 is-primary 1
|
||||
#status data1 is-replaying 0
|
||||
#status data1 is-syncing 0
|
||||
#status data1 open-count 1
|
||||
#status data1 replay_rate 0
|
||||
#status data1 allow-replay 1
|
||||
#status data1 attach 1
|
||||
#status data1 connect 1
|
||||
#status data1 sync 1
|
||||
UserParameter=system.mars[*], ( grep $1 /tmp/zabbix.mars || echo '0 0' ) | awk '{print $$2}'
|
||||
UserParameter=system.marsstatus[*], grep $1 /tmp/zabbix.mars | grep ' $2 ' | awk '{print $$4}' | head -n 1
|
||||
|
||||
|
|
|
|||
File diff suppressed because it is too large
Load Diff
Loading…
Reference in New Issue