Even though the statisticsserver can also run into oom errors, it should not consume a large share of the available memory. The statisticsserver is supposed to have a much smaller allocation limit compared to the indexserver. This following two configuration parameters are relevant:
statisticsserver.ini
-> memorymanager
allocationlimit
minallocationlimit
The allocation limit of the statisticsserver is either a percentage of the availably RAM or just the value of minallocationlimit, depending on what is the bigger value. In case of oom errors that indicate that the statisticsserver has reached its process allocation limit, it can make sense to increase the allocationlimit to 10% or 15% to have at least a workaround for the time being. However, further analysis regarding the heap memory usage has to be conducted (see above).
In the oom trace file, such a situation would look like this:
[3] PID=23840, SId=320543005, compactors active, alive, process name: hdbstatisticsse
AB=6755876864b (6.29gb), U=7135415505b (6.64gb), FSL=0b, PAL=6772669235b (6.30gb), TPAL=6772669235b (6.30gb), W=100
AB, PAL and TPAL are on the same values. Also in this case, further investigation regarding the involved pools is necessary, either using the information contained in the trace files or by using the following SQL statement:
SELECT TOP 15 MS.HOST, MS.SERVICE_NAME,MH.CATEGORY, MH.INCLUSIVE_SIZE_IN_USE, MH.EXCLUSIVE_SIZE_IN_USE FROM M_HEAP_MEMORY MH, M_SERVICES MS WHERE MH.PORT = MS.PORT AND MH.HOST = MS.HOST AND MS.SERVICE_NAME = 'statisticsserver' ORDER BY 4 DESC
statisticsserver.ini
-> memorymanager
allocationlimit
minallocationlimit
The allocation limit of the statisticsserver is either a percentage of the availably RAM or just the value of minallocationlimit, depending on what is the bigger value. In case of oom errors that indicate that the statisticsserver has reached its process allocation limit, it can make sense to increase the allocationlimit to 10% or 15% to have at least a workaround for the time being. However, further analysis regarding the heap memory usage has to be conducted (see above).
In the oom trace file, such a situation would look like this:
[3] PID=23840, SId=320543005, compactors active, alive, process name: hdbstatisticsse
AB=6755876864b (6.29gb), U=7135415505b (6.64gb), FSL=0b, PAL=6772669235b (6.30gb), TPAL=6772669235b (6.30gb), W=100
AB, PAL and TPAL are on the same values. Also in this case, further investigation regarding the involved pools is necessary, either using the information contained in the trace files or by using the following SQL statement:
SELECT TOP 15 MS.HOST, MS.SERVICE_NAME,MH.CATEGORY, MH.INCLUSIVE_SIZE_IN_USE, MH.EXCLUSIVE_SIZE_IN_USE FROM M_HEAP_MEMORY MH, M_SERVICES MS WHERE MH.PORT = MS.PORT AND MH.HOST = MS.HOST AND MS.SERVICE_NAME = 'statisticsserver' ORDER BY 4 DESC
No comments:
Post a Comment