911: My HANA One system on AWS has fallen and it can't get back up - Indexserver crash

Hi Juergen & Rahul,

I'm been busy getting my new instance up and running. Here are the trace specifics from indexserver_alert_hanaserver.trc:

[3084]{0}[0] 2013-01-17 19:45:09.598506 e Metadata ptl_shm.cc(00520) : ShmSystem::attach (shmid=12550492, align=67108864) - Cannot allocate memory

[3084]{0}[0] 2013-01-17 19:45:09.599115 e Row_Engine msglog.cc(00088) : Error during RowStore recovery: transaction rolled back due to unavailable resource (at ptime/storage/recovery/CheckpointMgr.cc:535 )

[3084]{0}[0] 2013-01-17 19:45:09.599273 e Basis Crash.cpp(00558) : Crash at /HDB/IMP/NewDB100_REL/src///sys/src/Basis/Diagnose/impl/FaultProtectionImpl.cpp:531

Reason:

exception 1: no.2100002 (Basis/Diagnose/impl/FaultProtectionImpl.cpp:531)

Illegal call to exit(), _exit() or _Exit() detected

exception throw location:

1: 0x00007f573f5b1f6a in exit_handler+0x46 at FaultProtectionImpl.cpp:531 (libhdbbasis.so)

2: 0x0000000000542bba in _exit+0x16 at LinuxMallocInitializer.cpp:143 (hdbindexserver)

3: 0x00007f5735f55897 in ptime::CheckpointMgr::restarter(void*)+0xd3 at CheckpointMgr.cc:536 (libhdbrskernel.so)

4: 0x00007f57354e9a00 in ptime::PtimeThread::run(void*)+0x10 at ptime_thread.h:131 (libhdbrskernel.so)

5: 0x00007f574c16c000 in TrexThreads::PoolThread::run()+0xc30 at PoolThread.cpp:255 (libhdbbasement.so)

6: 0x00007f574c16d7a8 in TrexThreads::PoolThread::run(void*&)+0x14 at PoolThread.cpp:104 (libhdbbasement.so)

7: 0x00007f573f641545 in Execution::Thread::staticMainImp(void**)+0x671 at Thread.cpp:448 (libhdbbasis.so)

8: 0x00007f573f64170d in Execution::Thread::staticMain(void*)+0x39 at Thread.cpp:512 (libhdbbasis.so)

[3084]{0}[0] 2013-01-17 19:45:09.645464 e Basis FaultProtectionImpl.cpp(00961) : SIGNAL 6 (SIGABRT) caught, sender PID: 3042, PID: 3042, thread: 2710[thr=3084]: Checkpointer, value int: 464540312, ptr: 0xffff880f1bb05298, time: 2013-01-17 19:45:09 000 Local

Instance HDB/00, OS Linux hanaserver 2.6.32.27-0.2-default #1 SMP 2010-12-29 15:03:02 +0100 x86_64

The scenario occurred when I was attempting to load up my test database with approx 30GB of data into ROW tables on the 63 GB instance of HANA One. The import of two of the tables took forever, so I stopped the instance and restarted it. I then learned about ALTER SYSTEM LOGGING OFF; However, the damage was done. I believe that I managed to get too much data into the ROW tables and HANA ran out of memory. Here are the size of the data files for the database:

hanaserver:/hanadata/HDB/data/mnt00001> ls -l ./hdb00001

total 24668

-rw------- 1 hdbadm sapsys 335577088 2013-01-17 22:43 datavolume_0000.dat

-rw-rw-r-- 1 hdbadm sapsys 36 2013-01-17 19:43 landscape.id

hanaserver:/hanadata/HDB/data/mnt00001> ls -l ./hdb00002

total 34172160

-rw------- 1 hdbadm sapsys 35130195968 2013-01-16 02:48 datavolume_0000.dat

hanaserver:/hanadata/HDB/data/mnt00001> ls -l ./hdb00003

total 96152

-rw------- 1 hdbadm sapsys 351059968 2013-01-16 02:03 datavolume_0000.dat

hanaserver:/hanadata/HDB/data/mnt00001> ls -l ./hdb00004

total 26444

-rw------- 1 hdbadm sapsys 271745024 2013-01-16 02:03 datavolume_0000.dat

The instance seems to be running, but the SapService isn't running so HANA Studio can connect to even get the diagnostic traces. I have to go to Linux. I say this because when I issue the ./HDB stop command, it takes about 6 minutes to shut down.

Is there any way to delete tables or truncate tables in a database so that I can recover it?

Thanks,

Bill

Interestingly this wasn't our problem, it got resolved by .

The ACTIVE_TABLES table had a corrupted delta and mergedog would try to merge it around 15 minutes after startup. At this point the indexserver would restart.

Lloyd found it by starting the indexserver in console mode and watching the logs. He then copied the affected table, deleted the original and copied it back.

Then it was fine. Thanks for the reply.

Hi Peter,

You are quite right on the memory limits - especially in the case of row tables. I ended up with a scenario where the data loader added too much data to the system and ideally should have thrown an error during the load. It did - just too late. The row table ended up having more data than could load into memory and the server crashed. Unfortunately, I could never recover other than tearing down the instance and building a new instance - this time using column tables.

Regards,

Bill