The current develop branch fails with test3 and test4, although test4 seems like it is reacting to the issues from test3. Here is what test3 failed with (there are 20 of these messages, one from each wtperf worker):
update failed for: 0000000000000000000000000000000106451787, range: 499999979 Error: WT_ERROR: non-specific WiredTiger error
[1395169858:339666][22032:00f77fa2e07f0000], file:test-003402.lsm: test-003402.lsm read error: failed to read 4096 bytes at offset 27713536: WT_ERROR: non-specific WiredTiger error
And that LSM file is only about 16Mb, so it is true that offset 27Mb is an error. However, it does not have a bloom filter yet:
-rw-rw-r-- 1 sue sue 2310144 Mar 18 19:11 test-003399.bf -rw-rw-r-- 1 sue sue 224284672 Mar 18 19:10 test-003399.lsm -rw-rw-r-- 1 sue sue 36864 Mar 18 19:11 test-003400.bf -rw-rw-r-- 1 sue sue 27721728 Mar 18 19:10 test-003400.lsm -rw-rw-r-- 1 sue sue 16039936 Mar 18 19:10 test-003402.lsm -rw-rw-r-- 1 sue sue 77824 Mar 18 19:11 test-003403.bf -rw-rw-r-- 1 sue sue 3084288 Mar 18 19:11 test-003403.lsm -rw-rw-r-- 1 sue sue 81920 Mar 18 19:11 test-003404.lsm
It appears that the code is attempting to access that chunk before it is ready.