-
Type:
Bug
-
Resolution: Done
-
Priority:
Major - P3
-
Affects Version/s: None
-
Component/s: WiredTiger
-
Labels:
-
Fully Compatible
-
ALL
With rc5, WT b-tree and the zlib block compressor there is either a memory leak or WT uses much more memory compared to rc4.
My 144G test server gets OOM after insert 37M docs and vsz, rss for mongod are 175590976 142382556 (175B, 142B) at that point. With rc4 vsz, rss were 100881768 92328172 (100B, 92B) after 400M inserts.
The test is run as described in SERVER-16886
Looking at /proc/smaps there is one huge allocation:
7f0fcd400000-7f297c000000 rw-p 00000000 00:00 0 Size: 107720704 kB Rss: 90840720 kB Pss: 90840720 kB Shared_Clean: 0 kB Shared_Dirty: 0 kB Private_Clean: 0 kB Private_Dirty: 90840720 kB Referenced: 90840080 kB Anonymous: 90840720 kB AnonHugePages: 0 kB Swap: 0 kB KernelPageSize: 4 kB MMUPageSize: 4 kB Locked: 0 kB VmFlags: rd wr mr mw me ac ??
And one other large allocation
7f2990000000-7f3999800000 rw-p 00000000 00:00 0 Size: 67264512 kB Rss: 207524 kB Pss: 207524 kB Shared_Clean: 0 kB Shared_Dirty: 0 kB Private_Clean: 0 kB Private_Dirty: 207524 kB Referenced: 190240 kB Anonymous: 207524 kB AnonHugePages: 0 kB Swap: 106344 kB KernelPageSize: 4 kB MMUPageSize: 4 kB Locked: 0 kB VmFlags: rd wr mr mw me ac ??
From log, I don't set the cache size, but it used 70G
2015-01-16T09:06:10.560-0800 I STORAGE [initandlisten] wiredtiger_open config: create,cache_size=70G,session_max=20000,eviction=(threads_max=4),statistics=(fast),log=(enabled=true,archive=true,path=journal,compressor=snappy),checkpoint=(wait=60,log_size=2GB),statistics_log=(wait=0),
- is depended on by
-
-
- Closed
-
- is related to
-
WT-1869 cache: maximum page size at eviction exceeds memory_page_max by a large factor
-
- Closed
-
- related to
-
-
- Closed
-
-
-
- Closed
-