[ruby-core:26336] [Bug #2283] Ruby 1.9.1p243 spinning with 100% CPU; perhaps rb_str_slice_bang-related

Bug #2283: Ruby 1.9.1p243 spinning with 100% CPU; perhaps rb_str_slice_bang-related
http://redmine.ruby-lang.org/issues/show/2283

Author: Mark Aiken
Status: Open, Priority: Normal
ruby -v: 1.9.1p243

I have a loop like this in some server code; the intent is to prune back a not-necessarily-ASCII string, character by character, until it fits into a certain number of *bytes*, without severing any multibyte characters rudely in the middle:

cur_length = mystring.bytesize

while cur_length > MAX_SIZE
  mystring.slice!(-1)
  cur_length = mystring.bytesize
end

Our server periodically seizes up with the Ruby workers grabbing 100% of CPU and becoming unresponsive. I attached gdb and dumped some backtraces of what the process is doing when it gets into this state; two such traces are attached. rb_str_slice_bang shows up frequently (but not always) in the backtraces; I have included a trace where it does *not* appear.

From the fact that rb_str_slice_bang usually appears, and given the type of request being processed by our server when this occurs, I suspect this string-trimming loop is causing the runaway process somehow. From the backtraces it looks like the process is furiously allocating new strings, then garbage-collecting them. The runaway processes do *NOT* bloat up in terms of memory consumption, though.

Is there anything intrinsically wrong with trying to trim the string in this way? Is this a known Ruby 1.9.1 issue? How can I help track it down?


----------------------------------------
http://redmine.ruby-lang.org