Details
-
Bug
-
Resolution: Fixed
-
Critical
-
3.0
-
Security Level: Public
-
None
-
OS X Mavericks 10.9.3
MacBook Pro Haswell, 16GB, 8 hyperthreaded cores.
3.0.0 from top-of-trunk; see attached manifest for specific versions.
Also occurs on Linux (Xeon, 24 cores)
-
Triaged
-
MacOSX 64-bit
-
Unknown
-
June 30 - July 18
Description
When running an updated version of revAB sim (append-heavy workload) I noticed very large spikes in memory during the workload - from ~200MB up to ~380MB (then back down again). This doesn't occur with 2.5.1 - see attached screenshots
Note that the "steady-state" memory usage between the two is similar - around 220MB when the test completes, but the spikes are pretty massive.
Steps to reproduce:
Single node, 1 bucket with 1024MB Server Quota, default settings.
Run the attached revAB_sim_v3.zip (note this is a little evolved from my last version of it):
unzip revAB_sim_v3.zip
./revAB_sim.py
Test takes ~1m on my laptop, using 8 client worker threads (CouchbaseClient objects).
Interestingly, I see a similar (but different-shaped) graph for 4 client threads - see attached. (change the 'WORKERS' variable in the script to reproduce.