3137: Refactor stats to record keep & fuse operations as well as bytes.

diff --git a/sdk/python/arvados/keep.py b/sdk/python/arvados/keep.py
index 346147e1be9e5d77a8752d061f5e1b67f151657f..0d5eeb37fe225313e7731b6d59dc44c2f6c8a490 100644 (file)
--- a/sdk/python/arvados/keep.py
+++ b/sdk/python/arvados/keep.py
@@ -669,6 +669,8 @@ class KeepClient(object):
         self._user_agent_pool = Queue.LifoQueue()
         self.upload_counter = Counter()
         self.download_counter = Counter()
+        self.put_counter = Counter()
+        self.get_counter = Counter()
 
         if local_store:
             self.local_store = local_store
@@ -875,6 +877,9 @@ class KeepClient(object):
         """
         if ',' in loc_s:
             return ''.join(self.get(x) for x in loc_s.split(','))
+
+        self.get_counter.add(1)
+
         locator = KeepLocator(loc_s)
         slot, first = self.block_cache.reserve_cache(locator.md5sum)
         if not first:
@@ -980,6 +985,8 @@ class KeepClient(object):
         elif not isinstance(data, str):
             raise arvados.errors.ArgumentError("Argument 'data' to KeepClient.put is not type 'str'")
 
+        self.put_counter.add(1)
+
         data_hash = hashlib.md5(data).hexdigest()
         loc_s = data_hash + '+' + str(len(data))
         if copies < 1:

diff --git a/services/fuse/arvados_fuse/__init__.py b/services/fuse/arvados_fuse/__init__.py
index fc7aace9194b44781fb11680ad6d3b954f588bd2..fd25aa9b5eacab7e4b6038df50387375147513c6 100644 (file)
--- a/services/fuse/arvados_fuse/__init__.py
+++ b/services/fuse/arvados_fuse/__init__.py
@@ -326,6 +326,8 @@ class Operations(llfuse.Operations):
 
         self.read_counter = arvados.keep.Counter()
         self.write_counter = arvados.keep.Counter()
+        self.read_ops_counter = arvados.keep.Counter()
+        self.write_ops_counter = arvados.keep.Counter()
 
         self.events = None
 
@@ -488,6 +490,8 @@ class Operations(llfuse.Operations):
     @catch_exceptions
     def read(self, fh, off, size):
         _logger.debug("arv-mount read %i %i %i", fh, off, size)
+        self.read_ops_counter.add(1)
+
         if fh in self._filehandles:
             handle = self._filehandles[fh]
         else:
@@ -503,6 +507,8 @@ class Operations(llfuse.Operations):
     @catch_exceptions
     def write(self, fh, off, buf):
         _logger.debug("arv-mount write %i %i %i", fh, off, len(buf))
+        self.write_ops_counter.add(1)
+
         if fh in self._filehandles:
             handle = self._filehandles[fh]
         else:

diff --git a/services/fuse/bin/arv-mount b/services/fuse/bin/arv-mount
index d15553456a2238fcee7d32d926517ada43d05785..96116a8ee0053e0ff613f2aa330830ec06cfb1d5 100755 (executable)
--- a/services/fuse/bin/arv-mount
+++ b/services/fuse/bin/arv-mount
@@ -16,6 +16,61 @@ import arvados.keep
 
 logger = logging.getLogger('arvados.arv-mount')
 
+class Stat(object):
+    def __init__(self, prefix, interval,
+                 egr_name, ing_name,
+                 egr_func, ing_func):
+        self.prefix = prefix
+        self.interval = interval
+        self.egr_name = egr_name
+        self.ing_name = ing_name
+        self.egress = egr_func
+        self.ingress = ing_func
+        self.egr = self.egress()
+        self.ing = self.ingress()
+
+    def update(self):
+        self.egr_prev = self.egr
+        self.ing_prev = self.ing
+        self.egr = self.egress()
+        self.ing = self.ingress()
+
+        delta = " -- interval %.4f seconds %d %s %d %s" % (self.interval,
+                                                           self.egr-self.egr_prev,
+                                                           self.egr_name,
+                                                           self.ing-self.ing_prev,
+                                                           self.ing_name)
+
+        sys.stderr.write("crunchstat: %s %d %s %d %s%s\n" % (self.prefix,
+                                                             self.egr,
+                                                             self.egr_name,
+                                                             self.ing,
+                                                             self.ing_name,
+                                                             delta))
+
+def statlogger(keep, ops):
+    interval = 10
+    calls = Stat("keepcalls", interval, "put", "get",
+                 keep.put_counter.get,
+                 keep.get_counter.get)
+    net = Stat("net:keep0", interval, "tx", "rx",
+               keep.upload_counter.get,
+               keep.download_counter.get)
+    fuseops = Stat("fuseops", interval,"write", "read",
+                   ops.write_ops_counter.get,
+                   ops.read_ops_counter.get)
+    blk = Stat("blkio:0:0", interval, "write", "read",
+               ops.write_counter.get,
+               ops.read_counter.get)
+
+    while True:
+        time.sleep(interval)
+        calls.update()
+        net.update()
+        fuseops.update()
+        blk.update()
+
+
 if __name__ == '__main__':
     # Handle command line parameters
     parser = argparse.ArgumentParser(
@@ -55,6 +110,8 @@ with "--".
     parser.add_argument('--read-only', action='store_false', help="Mount will be read only (default)", dest="enable_write", default=False)
     parser.add_argument('--read-write', action='store_true', help="Mount will be read-write", dest="enable_write", default=False)
 
+    parser.add_argument('--stats', action='store_true', help="Write stats to stderr", default=False)
+
     parser.add_argument('--exec', type=str, nargs=argparse.REMAINDER,
                         dest="exec_args", metavar=('command', 'args', '...', '--'),
                         help="""Mount, run a command, then unmount and exit""")
@@ -101,6 +158,11 @@ with "--".
         api = ThreadSafeApiCache(apiconfig=arvados.config.settings(),
                                  keep_params={"block_cache": arvados.keep.KeepBlockCache(args.file_cache)})
 
+        if args.stats:
+            statsthread = threading.Thread(target=statlogger, args=(api.keep, operations))
+            statsthread.daemon = True
+            statsthread.start()
+
         usr = api.users().current().execute(num_retries=args.retries)
         now = time.time()
         dir_class = None