"Exceeded soft private memory limit" issue breaks a job

[MeLight]

At a certain point of the mapreduce job on one of the worker_callbacks (/mapreduce/worker_callback/15734955148708B76E8DE-1) I'm getting this error:

Exceeded soft private memory limit of 128 MB with 128 MB after servicing 1261 requests total

And after that on each mapreduce request I get this error:

Traceback (most recent call last):
  File "/base/data/home/runtimes/python27/python27_lib/versions/1/google/appengine/runtime/wsgi.py", line 240, in Handle
    handler = _config_handle.add_wsgi_middleware(self._LoadHandler())
  File "/base/data/home/runtimes/python27/python27_lib/versions/1/google/appengine/runtime/wsgi.py", line 299, in _LoadHandler
    handler, path, err = LoadObject(self._handler)
  File "/base/data/home/runtimes/python27/python27_lib/versions/1/google/appengine/runtime/wsgi.py", line 85, in LoadObject
    obj = __import__(path[0])
ImportError: No module named mapreduce

Until the job dies. Please advise

[tkaitchuck]

I'm not sure about the second error. (Surely it already loaded mapreduce at that point.)
However likely the instance size you are using has too little memory for what you are attempting to do in the mapper. It could either be, you are using more memory than you are aware of. (Perhaps because there are many reduce shards, and the write buffers are taking a lot of ram) Or perhaps your instance size is simply too small, (While it is possible, it is difficult to run an MR in under 128mb.)

[tkaitchuck]

Actually I didn't notice, the 1261 requests. That's a lot. So clearly it is not dying right away. Are those all MR requests? or are you also serving other traffic on the same module? I highly recommend putting MR jobs in their own module.