svc/connect.in: Report statistics about adopted peers.

These are thrown into the `info' output along with the configuration.

diff --git a/svc/connect.8.in b/svc/connect.8.in
index 7112fb0..0d951bf 100644 (file)
--- a/svc/connect.8.in
+++ b/svc/connect.8.in
@@ -523,7 +523,7 @@ provided about each peer, in the form of subsequent tokens.  Clients
 should be prepared to ignore such tokens.)
 .SP
 .BI "info " peer
-Lists the database record for the named
+Lists the database record and additional information about the named
 .IR peer .
 For each key/value pair, a line
 .RS
@@ -532,6 +532,67 @@ For each key/value pair, a line
 .IB key = value
 .PP
 is output.  The key/value pairs are output in an arbitrary order.
+.PP
+In addition to the fields of the peer's database record, the following
+additional keys are defined.
+.TP
+.B failures
+The number of failed pings in the current or most recent batch, in
+decimal.
+.TP
+.B last-ping
+The round-trip time of the most recent ping in milliseconds, in the form
+.IB nn.n ms\fR,
+or
+.B timeout
+if the most recent ping timed out,
+or
+.B \-
+if no pings have yet completed.
+.TP
+.B max-ping
+The maximum successful ping time so far in milliseconds, in the form
+.IB nn.n ms\fR,
+or
+.B \-
+if no pings have yet succeeded.
+.TP
+.B mean-ping
+The average successful ping time so far in milliseconds, in the form
+.IB nn.n ms\fR,
+or
+.B \-
+if no pings have yet succeeded.
+.TP
+.B min-ping
+The minimum successful ping time so far in milliseconds, in the form
+.IB nn.n ms\fR,
+or
+.B \-
+if no pings have yet succeeded.
+.TP
+.B n-lost
+The number of pings which have been declared timed out so far, in
+decimal.
+.TP
+.B n-ping
+The number of successful pings so far, in decimal.
+.TP
+.B sd-ping
+The standard deviation of ping times so far in milliseconds, in the form
+.IB nn.n ms\fR,
+or
+.B \-
+if no pings have yet succeeded.
+.TP
+.B state
+One of the strings:
+.B idle
+if the peer has responded to a ping recently, and we are waiting for the
+.B every
+delay before we try again; or
+.B check
+if we are currently waiting for a ping to return.
 .RE
 .SP
 .BI "kick " peer

diff --git a/svc/connect.in b/svc/connect.in
index 3e9fa0a..e898141 100644 (file)
--- a/svc/connect.in
+++ b/svc/connect.in
@@ -33,6 +33,7 @@ import tripe as T
 import os as OS
 import signal as SIG
 import errno as E
+from math import sqrt
 import cdb as CDB
 import mLib as M
 import re as RX
@@ -395,6 +396,12 @@ class PingPeer (object):
     me.seq = _pingseq
     _pingseq += 1
     me._failures = 0
+    me._last = '-'
+    me._nping = 0
+    me._nlost = 0
+    me._sigma_t = 0
+    me._sigma_t2 = 0
+    me._min = me._max = '-'
     if pingnow:
       me._timer = None
       me._ping()
@@ -464,18 +471,47 @@ class PingPeer (object):
       if stuff[0] == 'ping-ok':
         if me._failures > 0:
           S.warn('connect', 'ping-ok', me._peer)
+        t = float(stuff[1])
+        me._last = '%.1fms' % t
+        me._sigma_t += t
+        me._sigma_t2 += t*t
+        me._nping += 1
+        if me._min == '-' or t < me._min: me._min = t
+        if me._max == '-' or t > me._max: me._max = t
         me._timer = M.SelTimer(time() + me._every, me._time)
       elif stuff[0] == 'ping-timeout':
         me._failures += 1
+        me._nlost += 1
         S.warn('connect', 'ping-timeout', me._peer,
                'attempt', str(me._failures), 'of', str(me._retries))
         if me._failures < me._retries:
           me._ping()
+          me._last = 'timeout'
         else:
           me._reconnect()
       elif stuff[0] == 'ping-peer-died':
         me._pinger.kill(me._peer)
 
+  def info(me):
+    if not me._nping:
+      mean = sd = '-'
+    else:
+      mean = me._sigma_t/me._nping
+      sd = sqrt(me._sigma_t2/me._nping - mean*mean)
+    n = me._nping + me._nlost
+    if not n: pclost = '-'
+    else: pclost = '%d' % ((100*me._nlost + n//2)//n)
+    return { 'last-ping': me._last,
+             'mean-ping': '%.1fms' % mean,
+             'sd-ping': '%.1fms' % sd,
+             'n-ping': '%d' % me._nping,
+             'n-lost': '%d' % me._nlost,
+             'percent-lost': pclost,
+             'min-ping': '%.1fms' % me._min,
+             'max-ping': '%.1fms' % me._max,
+             'state': me._timer and 'idle' or 'check',
+             'failures': me._failures }
+
   @T._callback
   def _time(me):
     """
@@ -759,10 +795,16 @@ def cmd_info(name):
   """
   try: peer = Peer(name)
   except KeyError: raise T.TripeJobError('unknown-peer', name)
-  items = list(peer.list())
+  d = {}
+  try: pp = pinger.find(name)
+  except KeyError: pass
+  else: d.update(pp.info())
+  items = list(peer.list()) + d.keys()
   items.sort()
   for i in items:
-    T.svcinfo('%s=%s' % (i, peer.get(i)))
+    try: v = d[i]
+    except KeyError: v = peer.get(i)
+    T.svcinfo('%s=%s' % (i, v))
 
 def cmd_userpeer(user):
   """