Hello
MonetDB Nov2008 release, built with all standard options on RedHat EL 5 (2.6.18-92.el5 #1 SMP), on HP Proliant DL380 8-way, 16GB RAM / 16GB swap, with dbfarm on a RAID0 internal 7-disk array.
I'm doing bulk load tests (using mclient to run COPY INTO table FROM file), repeatedly loading 500,000 row (68M) CSV files into a fact table, while periodically running a suite of test queries in a separate process to test query times as the fact table grows.
The problem is that as the fact table grows to around 80Million records, the database starts to consume all available system memory and swap, the system thrashes and comes to its knees. I also get database corruption causing foreign key violations. (More details below)
Has anyone else seen this? Any ideas if it's a bug, memory leak? Or could this be caused by operator error? Suggestions?
Cheers Bob
PS More detailed account of the problem below:
Everything starts off fine, files load in 8-10 seconds, queries quite fast.
As the table starts to fill, the queries (not unexpectedly) start to slow. But (unexpectedly) the memory consumption of the mserver5 process steadily increases.
At some point, around 80Million records, the system performance rapidly starts to degrade. Query times dramatically increase, load time also dramatically increases, and the entire system thrashes and becomes unusable.
Running 'top' shows that mserver5 has consumed vurtually all physical memory, and most of the swap too.
top - 16:33:34 up 28 days, 23:52, 8 users, load average: 1.68, 1.51, 1.96 Tasks: 224 total, 1 running, 223 sleeping, 0 stopped, 0 zombie Cpu(s): 0.1%us, 1.5%sy, 0.0%ni, 85.9%id, 12.4%wa, 0.0%hi, 0.0%si, 0.0%st Mem: 16439196k total, 16298404k used, 140792k free, 2300k buffers Swap: 18876364k total, 13315960k used, 5560404k free, 9824352k cached PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND 32380 monetdb 18 0 54.6g 14g 9.4g S 12 94.9 107:19.46 mserver5 29217 bostr 15 0 290m 5552 2264 S 1 0.0 12:01.37 gnome-terminal
The first time this happened, I stopped the load scripts, restarted the database, and that helped things for a little while, but it wasn't long before the memory was all consumed again, and performance went to the dogs. Then loads started failing with foreign-key violation errors.. Inspection showed a corrupted dimension table that somehow ended up with a bunch of bad values in it..
So I blew away, and recreated the database.
This time I tried adding the line 'gdk_nr_threads=1' to the monetdb5.conf file - having seen a defect report related to 'COPY INTO ' being buggy and this being the solution. But the same thing happened again.. it ran fine again until I had about 100Million rows, then memory problems/thrashing. Restarted the server again, and shortly after the data loads failed again with foreign-key violations and a corrupted dimension table..