andy/glusterfs
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

gnfs-loganalyse: Track fh & entry operations given filename

Browse Source 
Introduces support for outputting only those NFS operations
that operate on a given filename and all the filehandles that
are seen in the log file for that filename.

Signed-off-by: Shehjar Tikoo <shehjart@gluster.com>
Signed-off-by: Anand Avati <avati@gluster.com>

BUG: 913 (NFS-specific log analyzer tool)
URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=913
This commit is contained in:
Shehjar Tikoo 2011-04-18 00:58:49 +00:00 committed by Anand Avati
parent 4ba7635f4d
commit ef3d2a48f4
1 changed files with 137 additions and 13 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

150
extras/gnfs-loganalyse.py
View File

@ -24,6 +24,13 @@ import sys
class NFSRequest:
def requestIsEntryOp (self):
op = self.op
if op == "CREATE" or op == "LOOKUP" or op == "REMOVE" or op == "LINK" or op == "RENAME" or op == "MKDIR" or op == "RMDIR" or op == "SYMLINK" or op == "MKNOD":
return 1
else:
return 0
def __init__ (self, logline, linecount):
self.calllinecount = 0
self.xid = ""
@ -31,8 +38,13 @@ class NFSRequest:
self.opdata = ""
self.replydata = ""
self.replylinecount = 0
self.timestamp = ""
self.entryname = ""
self.gfid = ""
self.replygfid = ""
tokens = logline.strip ().split (" ")
self.timestamp = tokens[0] + " " + tokens[1]
if "XID:" not in tokens:
return None
@ -47,6 +59,11 @@ class NFSRequest:
opidx = tokens.index ("args:")
self.op = tokens [opidx-1].strip (":")
self.opdata = " ".join(tokens [opidx+1:])
if self.requestIsEntryOp ():
nameidx = tokens.index ("name:")
self.entryname = tokens[nameidx + 1].strip (",")
gfididx = tokens.index ("gfid")
self.gfid = tokens[gfididx +1].strip(",")
def getXID (self):
@ -54,25 +71,56 @@ class NFSRequest:
def setReply (self, logline, linecount):
tokens = logline.strip ().split (" ")
timestamp = tokens[0] + " " + tokens[1]
statidx = tokens.index ("NFS:")
self.replydata = " ".join (tokens [statidx+1:])
self.replydata = " TimeStamp: " + timestamp + " " + " ".join (tokens [statidx+1:])
self.replylinecount = linecount
if "gfid" in tokens:
gfididx = tokens.index ("gfid")
self.replygfid = tokens [gfididx + 1].strip(",")
def dump (self):
print "ReqLine: " + str(self.calllinecount) + " XID: " + self.xid + " " + self.op + " ARGS: " + self.opdata + " RepLine: " + str(self.replylinecount) + " " + self.replydata
print "ReqLine: " + str(self.calllinecount) + " TimeStamp: " + self.timestamp + ", XID: " + self.xid + " " + self.op + " ARGS: " + self.opdata + " RepLine: " + str(self.replylinecount) + " " + self.replydata
class NFSLogAnalyzer:
def __init__ (self):
def __init__ (self, optn, trackfilename, tracknamefh, stats):
self.stats = stats
self.xid_request_map = {}
self.orphan_replies = {}
self.rqlist = []
self.CALL = 1
self.REPLY = 2
self.optn = optn
self.trackfilename = trackfilename
self.tracknamefh = tracknamefh
self.trackedfilehandles = []
def handle_call_line (self, logline, linecount):
newreq = NFSRequest (logline, linecount)
xid = newreq.getXID ()
self.xid_request_map [xid] = newreq
if (self.optn == SYNTHESIZE):
self.xid_request_map [xid] = newreq
self.rqlist.append(newreq)
elif self.optn == TRACKFILENAME:
if newreq.requestIsEntryOp():
if newreq.entryname == self.trackfilename:
self.xid_request_map [xid] = newreq
self.rqlist.append(newreq)
else:
del newreq
elif self.tracknamefh == ENABLE_TRACKNAME_FH:
if len (self.trackedfilehandles) > 0:
if newreq.gfid in self.trackedfilehandles:
self.xid_request_map [xid] = newreq
self.rqlist.append(newreq)
else:
del newreq
else:
del newreq
else:
del newreq
def handle_reply_line (self, logline, linecount):
tokens = logline.strip ().split (" ")
@ -82,7 +130,10 @@ class NFSLogAnalyzer:
if xid not in self.xid_request_map.keys ():
self.orphan_replies [xid] = logline
else:
self.xid_request_map [xid].setReply (logline, linecount)
rq = self.xid_request_map [xid]
rq.setReply (logline, linecount)
if rq.requestIsEntryOp() and rq.entryname == self.trackfilename:
self.trackedfilehandles.append (rq.replygfid)
def analyzeLine (self, logline, linecount):
tokens = logline.strip ().split (" ")
@ -102,43 +153,116 @@ class NFSLogAnalyzer:
self.handle_reply_line (logline, linecount)
def getStats (self):
if self.stats == 0:
return
rcount = len (self.xid_request_map.keys ())
orphancount = len (self.orphan_replies.keys ())
print "Requests: " + str(rcount) + ", Orphans: " + str(orphancount)
def dump (self):
self.getStats ()
for rq in self.xid_request_map.values ():
for rq in self.rqlist:
rq.dump ()
del rq
self.rqlist = []
self.orphan_replies = {}
self.xid_request_map = {}
linecount = 0
la = NFSLogAnalyzer ()
SYNTHESIZE = 1
TRACKFILENAME = 2
ENABLESTATS = 1
DISABLESTATS = 0
ENABLE_TRACKNAME_FH = 1
DISABLE_TRACKNAME_FH = 0
progmsgcount = 1000
dumpinterval = 2000000
dumpinterval = 200000
operation = SYNTHESIZE
stats = ENABLESTATS
tracknamefh = DISABLE_TRACKNAME_FH
trackfilename = ""
"""
Print the progress of the analysing operations every X number of lines read from
the logs, where X is the argument provided to this option.
Use this to print a status message every say 10000 lines processed or 100000
lines processed to know how much longer the processing will go on for.
USAGE: --progress <NUMLINES>
"""
if "--progress" in sys.argv:
idx = sys.argv.index ("--progress")
progmsgcount = int(sys.argv[idx+1])
"""
The replies for a NFS request can be separated by hundreds and even thousands
of other NFS requests and replies. These can be spread over many hundreds and
thousands of log lines. This script maintains a memory dict to map each request
to its reply using the XID. Because this is in-core, there is a limit to the
number of entries in the dict. At regular intervals, it dumps the mapped
requests and the replies into the stdout. The requests whose replies were not
found at the point of dumping are left as orphans, i.e. without info about the
replies. Use this option to tune the number of lines to maximize the number of
requests whose replies are found while balancing the dict size with memory
on the machine. The default works fine for most cases.
USAGE: --dump <NUMLINES>
"""
if "--dump" in sys.argv:
idx = sys.argv.index ("--dump")
dumpinterval = int(sys.argv[idx+1])
"""
The default operation of the script is to output all the requests mapped to
their replies in a single line. This operation mode can be changed by this
argument. It is used to print only those operations that were performed on the
filename given as the argument to this option. Only those entry operations are
printed which contain this filename.
USAGE: --trackfilename <filename>
"""
if "--trackfilename" in sys.argv:
idx = sys.argv.index ("--trackfilename")
trackfilename = sys.argv[idx + 1]
operation = TRACKFILENAME
"""
At every dump interval, some stats are printed about the dumped lines.
Use this option to disable printing that to avoid cluttering the
output.
"""
if "--nostats" in sys.argv:
stats = DISABLESTATS
"""
While tracking a file using --trackfilename, we're only given those
operations which contain the filename. This excludes a large number
of operations which operate on that file using its filehandle instead of
the filename. This option enables outputting those operations also. It
tracks every single file handle that was ever seen in the log for a given
filename.
USAGE: --trackfilename
"""
if "--tracknamefh" in sys.argv:
tracknamefh = ENABLE_TRACKNAME_FH
la = NFSLogAnalyzer (operation, trackfilename, tracknamefh, stats)
for line in sys.stdin:
linecount = linecount + 1
if linecount % dumpinterval == 0:
sys.stderr.write ("Dumping data..\n")
la.dump ()
del la
la = NFSLogAnalyzer ()
if linecount % progmsgcount == 0:
sys.stderr.write ("Integrating line: "+ str(linecount) + "\n")
la.analyzeLine (line, linecount)
la.dump ()