diff options
Diffstat (limited to 'pgaas/src/stage/opt/app/pgaas/bin/check_cluster')
| -rwxr-xr-x | pgaas/src/stage/opt/app/pgaas/bin/check_cluster | 28 |
1 files changed, 22 insertions, 6 deletions
diff --git a/pgaas/src/stage/opt/app/pgaas/bin/check_cluster b/pgaas/src/stage/opt/app/pgaas/bin/check_cluster index 0af6d95..c55d973 100755 --- a/pgaas/src/stage/opt/app/pgaas/bin/check_cluster +++ b/pgaas/src/stage/opt/app/pgaas/bin/check_cluster @@ -23,7 +23,7 @@ # -v verbose # # DESCRIPTION -# Loop through the nodes in the cluster, using pgwget to determine how many are masters, secondaries, or not up. +# Loop through the nodes in the cluster, using pgwget to determine how many are masters, secondaries, down for maintenance, or not up. # Complain about certain situations. # If there are multiple masters, and this not the first master in the list, then: # run pg_ctl_restart @@ -44,19 +44,22 @@ usage() { exec 1>&2 [ $# -gt 0 ] && echo "$@" - echo "Usage: $0 [-v] [-l] [-t timeout]" + echo "Usage: $0 [-v] [-l] [-t timeout] [-d file]" echo -e " -l do not check localhost first (and restarting the service if necessary)" echo -e " -t timeout set how long to wait when accessing the servers" echo -e " -v verbose" + echo -e " -d file duplicate the status output to the given file" exit 1 } VERBOSE=false TIMEOUT=10 TESTLOCAL=: -while getopts lt:v c +DFILE= +while getopts d:lt:v c do case "$c" in + d ) DFILE=$OPTARG ;; l ) TESTLOCAL=false ;; t ) TIMEOUT=$OPTARG ;; v ) VERBOSE=: ;; @@ -70,13 +73,16 @@ master_count=0 secondary_count=0 total_count=0 down_count=0 +maintenance_count=0 -DOWNS= MASTERS= SECONDARIES= +MAINTENANCES= +DOWNS= MSEP= SSEP= +BSEP= DSEP= HOSTNAME=$(hostname -f) FOUNDPREVIOUSMASTER= @@ -86,7 +92,7 @@ if $TESTLOCAL then isrw=`pgwget --tries=1 --read-timeout=$TIMEOUT --quiet -O/dev/stdout http://localhost:8000/isrw` case "$isrw" in - Master | Secondary ) ;; + Master | Secondary | Maintenance ) ;; * ) echo "$(date)|WARNING|RESTARTED|Local iDNS-responder.py not responding. Restarting." ps -fu postgres | grep "python3 /opt/app/postgresql-prep/bin/iDNS-responder.py" | grep -v grep | awk '{print "kill " $2}' | sh @@ -118,6 +124,12 @@ do SECONDARIES="$SECONDARIES$SSEP$i" SSEP=" " ;; + Maintenance ) + (( maintenance_count = maintenance_count + 1 )) + (( total_count = total_count + 1 )) + MAINTENANCES="$MAINTENANCES$BSEP$i" + BSEP=" " + ;; * ) DOWNS="$DOWNS$DSEP$i" DSEP=" " @@ -130,7 +142,11 @@ done (( up_count = master_count + secondary_count )) date=$(date) -echo "$date|INFO|masters=$master_count $MASTERS|secondaries=$secondary_count $SECONDARIES|down=$down_count $DOWNS|" +output="$date|INFO|masters=$master_count $MASTERS|secondaries=$secondary_count $SECONDARIES|maintenance=$maintenance_count $MAINTENANCES|down=$down_count $DOWNS|" +echo "$output" +if [ -n "$DFILE" ] +then (umask 022; echo "$output" > $DFILE.tmp && mv $DFILE.tmp $DFILE) +fi FORCEROOFF=/var/run/postgresql/force-ro-off if [ $master_count -lt 1 ] |