ceph/doc/dev/logs.rst

   1 ============
   2  Debug logs
   3 ============
   4
   5 The main debugging tool for Ceph is the dout and derr logging functions.
   6 Collectively, these are referred to as "dout logging."
   7
   8 Dout has several log faculties, which can be set at various log
   9 levels using the configuration management system. So it is possible to enable
  10 debugging just for the messenger, by setting debug_ms to 10, for example.
  11
  12 The dout macro avoids even generating log messages which are not going to be
  13 used, by enclosing them in an "if" statement. What this means is that if you
  14 have the debug level set at 0, and you run this code::
  15
  16         dout(20) << "myfoo() = " << myfoo() << dendl;
  17
  18
  19 myfoo() will not be called here.
  20
  21 Unfortunately, the performance of debug logging is relatively low. This is
  22 because there is a single, process-wide mutex which every debug output
  23 statement takes, and every debug output statement leads to a write() system
  24 call or a call to syslog(). There is also a computational overhead to using C++
  25 streams to consider. So you will need to be parsimonious in your logging to get
  26 the best performance.
  27
  28 Sometimes, enabling logging can hide race conditions and other bugs by changing
  29 the timing of events. Keep this in mind when debugging.
  30
  31 Performance counters
  32 ====================
  33
  34 Ceph daemons use performance counters to track key statistics like number of
  35 inodes pinned. Performance counters are essentially sets of integers and floats
  36 which can be set, incremented, and read using the PerfCounters API.
  37
  38 A PerfCounters object is usually associated with a single subsystem.  It
  39 contains multiple counters. This object is thread-safe because it is protected
  40 by an internal mutex. You can create multiple PerfCounters objects.
  41
  42 Currently, three types of performance counters are supported: u64 counters,
  43 float counters, and long-run floating-point average counters. These are created
  44 by PerfCountersBuilder::add_u64, PerfCountersBuilder::add_fl, and
  45 PerfCountersBuilder::add_fl_avg, respectively. u64 and float counters simply
  46 provide a single value which can be updated, incremented, and read atomically.
  47 floating-pointer average counters provide two values: the current total, and
  48 the number of times the total has been changed. This is intended to provide a
  49 long-run average value.
  50
  51 Performance counter information can be read in JSON format from the
  52 administrative socket (admin_sock). This is implemented as a UNIX domain
  53 socket. The Ceph performance counter plugin for collectd shows an example of how
  54 to access this information. Another example can be found in the unit tests for
  55 the administrative sockets.