SAP HANA StatisticsServer Out of Memory Dump on Application

[Former Member]

Hello All,

We are on HANA SPS 08 Rev 85, Statistics server(standalone) was restarted due to OOM. My question here is does this cause a dump(ST22) in XXX(SAP Application system)?

XXX(SAP Application) ------> YYY(SLT) -----> HANA

Does restarting/stopping statistics server causes system halt? Will there be any effect on HANA during statisticsserver restart?

Any information is appreciated.

Regards,

Dheeraj

[Former Member · Accepted solution]

We have again and again seen a global impact of OOM situations in the standalone statistics server - for example distributed transactions from indexserver to statistics server were impacted and so also the indexserver suffered. So I wouldn't exclude that there can be also dumps in ST22 that are finally caused by a statistics server problem. In any case I highly recommend to migrate to ESS, because this will have an overall stabilizing impact. See SAP Note 2147247 for more information.

[Former Member · Accepted solution]

Thanks Martin for reply.

How / why distributed transactions are run between indexserver and statistics server.

I couldn't find anything related to statistics server impact on indexserver. Any documentation on this?

Also, we have below error on HANA. How does these below error effect other components?

Indexserver:

2015-06-23 09:50:03.254502 i TraceContext     TraceContext.cpp(00721) : UserName=_SYS_STATISTICS

[4779]{300320}[-1/-1] 2015-06-23 09:50:03.254491 i SQLSessionCmd    Statement.cc(03380) : session control command is performed by 300320, user=_SYS_STATISTICS, application user=, application source=, query=ALTER SYSTEM DISCONNECT SESSION '222606'

Statisticsserver:

_SYS_STATISTICS.STATISTICS_LASTVALUES(host,port,name,index,reached_at,value) values ('*', 0, ?, ?, ?, ?) where name=? and index=?', user: '_SYS_STATISTICS', called from 'execte_local aborted'

[61474]{-1}[-1/-1] 2015-06-23 09:51:13.346665 e StatsServ        ResourceGuard.cpp(00020) : Exception caught while cleaning-up resource:

exception  1: no.9002001 (StatisticsServerAsSQLClient/connection/OKAssurer.hpp:93)

      SQLDBC error -10821: Session not connected. SQLState: .

Regards,
Dheeraj

[Former Member · Accepted solution]

Every query on statistics server tables (e.g. in order to evaluate history information) is executed in the indexserver, but executed in the statisticsserver. So it is a distributed transaction. Errors like the -10821 issue you mentioned can become critical - e.g. if some hanging transactions remain for that reason. By switching to ESS you rule out all of these problem scenarios.

[Former Member · Accepted solution]

Incident(531500) has been created for this SQLDBC error -10821, can you please help on this.

I don't understand how statistics server effects indexserver as it only resides in master node, other nodes will have enough memory to perform tasks.

In the above post even agrees that no other service is interrupted.

Regards,

Dheeraj

[Former Member · Accepted solution]

I am busy with other tasks, so I can't have a look at the SAP incident. When a distributed transaction runs into an error and isn't cleaned up properly (e.g. due to a bug), various general problems are possible due to MVCC, garbage collection, transactional lock waits and internal lock waits. This may be an exception, but I have seen it more than once.

[Former Member]

does this cause a dump(ST22) in XXX(SAP Application system)? - > I wont cause a dump in SAP application server.

XXX(SAP Application) ------> YYY(SLT) -----> HANA

Does restarting/stopping statistics server causes system halt? -> SAP application will not hault.

Will there be any effect on HANA during statisticsserver restart? -> I belive there will be no effect. HANA will not be able to collect the statistics.

Regards,

Pavan Gunda

[Former Member]

Hi Dheeraj,

We had a few issues around HANA Statistics server(standalone) before the Embedded Statistics Server(ESS) migration

We have a Scale out system for BW on HANA

I do not think the restart of HANA Statistics server caused a dump anytime or impacted the entire HANA system to go into a hung state, unless the statistics server was using up the entire memory on the system. In our case due to the scale out setup we had enough memory on the systems so we never saw such a scenario

Due to the issues we noticed we had to eventually migrate to ESS and it is much more stable

In your case especially on SAP SLT HANA theoretically it should not be impacting the SLT replication except for the monitoring and alerts generation would not work until the restart is completed. I would suggest to test it in Non prod first before trying it out on Production HANA system

Hope it helps

Sunil

[Former Member]

Not sure if you'll see a dump, but will see a log in HANA and you can setup monitoring alerts.

The other services will continue to run if the statistics server bounces.

You should put steps in place to prevent the statistic server from getting OOMs.

1929538 - HANA Statistics Server - Out of Memory

-Jake

[Former Member]

Hello Jake,

Thanks for the reply.

Yes, I agree with you that all other services continue to run. But does statistics server interrupts connections between different systems? ex: SAP SLT <----> HANA

Regards,

Dheeraj