Synnefo » snf-ganeti » ganeti-local

	doc/design-node-state-cache.rst \

   design-node-state-cache.rst

================

Node State Cache

================

.. contents:: :depth: 4

This is a design doc about the optimization of machine info retrieval.

Current State

=============

Currently every RPC call is quite expensive as a TCP handshake has to be

made as well as SSL negotiation. This especially is visible when getting

node and instance info over and over again.

This data, however, is quite easy to cache but needs some changes to how

we retrieve data in the RPC as this is spread over several RPC calls

and are hard to unify.

Proposed changes

================

To overcome this situation with multiple information retrieval calls we

introduce one single RPC call to get all the info in a organized manner,

for easy store in the cache.

As of now we have 3 different information RPC calls:

- ``call_node_info``: To retrieve disk and hyper-visor information

- ``call_instance_info``: To retrieve hyper-visor information for one

  instance

- ``call_all_instance_info``: To retrieve hyper-visor information for

  all instances

Not to mention that ``call_all_instance_info`` and

``call_instance_info`` return different information in the dict.

To unify the data and organize them we introduce a new RPC call

``call_node_snapshot`` doing all of the above in one go. Which

data we want to know will be specified about a dict of request

types: CACHE_REQ_HV, CACHE_REQ_DISKINFO, CACHE_REQ_BOOTID

As this cache is representing the state of a given node we use the

name of a node as the key to retrieve the data from the cache. A

name-space separation of node and instance data is not possible at the

current point. This is due to the fact that some of the node hyper-visor

information like free memory is correlating with instances running.

An example of how the data for a node in the cache looks like::

  {

    constants.CACHE_REQ_HV: {

      constants.HT_XEN_PVM: {

        _NODE_DATA: {

          "memory_total": 32763,

          "memory_free": 9159,

          "memory_dom0": 1024,

          "cpu_total": 4,

          "cpu_sockets": 2

        },

        _INSTANCES_DATA: {

          "inst1": {

            "memory": 4096,

            "state": "-b----",

            "time": 102399.3,

            "vcpus": 1

          },

          "inst2": {

            "memory": 4096,

            "state": "-b----",

            "time": 12280.0,

            "vcpus": 3

          }

        }

      }

    },

    constants.CACHE_REQ_DISKINFO: {

      "xenvg": {

        "vg_size": 1048576,

        "vg_free": 491520

      },

    }

    constants.CACHE_REQ_BOOTID: "0dd0983c-913d-4ce6-ad94-0eceb77b69f9"

  }

This way we get easy to organize information which can simply be arranged in

the cache.

The 3 RPC calls mentioned above will remain for compatibility reason but

will be simple wrappers around this RPC call.

Cache invalidation

------------------

The cache is invalidated at every RPC call which is not proven to not

modify the state of a given node. This is to avoid inconsistency between

cache and actual node state.

There are some corner cases which invalidates the whole cache at once as

they usually affect other nodes states too:

 - migrate/failover

 - import/export

A request will be served from the cache if and only if it can be

fulfilled entirely from it (i.e. all the CACHE_REQ_* entries are already

present). Otherwise, we will invalidate the cache and actually do the

remote call.

In addition, every cache entry will have a TTL of about 10 minutes which

should be enough to accommodate most use cases.

We also allow an option to the calls to bypass the cache completely and

do a force remote call. However, this will invalidate the present

entries and populate the cache with the new retrieved values.

Additional cache population

---------------------------

Besides of the commands which calls above RPC calls, a full cache

population can also be done by a separate new op-code run by

``ganeti-watcher`` periodically. This op-code will be used instead of

the old ones.

Possible regressions

====================

As we change from getting "one hyper-visor information" to "get all we

know about this hyper-visor"-style we have a regression in time of

execution. The execution time is about 1.8x more in process execution

time. However, this does not include the latency and negotiation time

needed for each separate RPC call. Also if we hit the cache all 3 costs

will be 0. The only time taken is to look up the info in the cache and

the deserialization of the data. Which takes down the time from today

~300ms to ~100ms.

.. vim: set textwidth=72 :

.. Local Variables:

.. mode: rst

.. fill-column: 72

.. End:

#

#

# Copyright (C) 2011 Google Inc.

#

# This program is free software; you can redistribute it and/or modify

# it under the terms of the GNU General Public License as published by

# the Free Software Foundation; either version 2 of the License, or

# (at your option) any later version.

#

# This program is distributed in the hope that it will be useful, but

# WITHOUT ANY WARRANTY; without even the implied warranty of

# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU

# General Public License for more details.

#

# You should have received a copy of the GNU General Public License

# along with this program; if not, write to the Free Software

# Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA

# 02110-1301, USA.

"""This module implements caching."""

import time

from ganeti import locking

from ganeti import serializer

TIMESTAMP = "timestamp"

TTL = "ttl"

VALUE = "value"

class CacheBase:

  """This is the base class for all caches.

  """

  def __init__(self):

    """Base init method.

    """

  def Store(self, key, value, ttl=0):

    """Stores key with value in the cache.

    @param key: The key to associate this cached value

    @param value: The value to cache

    @param ttl: TTL in seconds after when this entry is considered outdated

    @returns: L{True} on success, L{False} on failure

    """

    raise NotImplementedError

  def GetMulti(self, keys):

    """Retrieve multiple values from the cache.

    @param keys: The keys to retrieve

    @returns: The list of values

    """

    raise NotImplementedError

  def Get(self, key):

    """Retrieve the value from the cache.

    @param key: The key to retrieve

    @returns: The value or L{None} if not found

    """

    raise NotImplementedError

  def Invalidate(self, keys):

    """Invalidate given keys.

    @param keys: The list of keys to invalidate

    @returns: L{True} on success, L{False} otherwise

    """

    raise NotImplementedError

  def Flush(self):

    """Invalidates all of the keys and flushes the cache.

    """

    raise NotImplementedError

  def ResetState(self):

    """Used to reset the state of the cache.

    This can be used to reinstantiate connection or any other state refresh

    """

  def Cleanup(self):

    """Cleanup the cache from expired entries.

    """

class SimpleCache(CacheBase):

  """Implements a very simple, dict base cache.

  """

  CLEANUP_ROUND = 1800

  _LOCK = "lock"

  def __init__(self, _time_fn=time.time):

    """Initialize this class.

    @param _time_fn: Function used to return time (unittest only)

    """

    CacheBase.__init__(self)

    self._time_fn = _time_fn

    self.cache = {}

    self.lock = locking.SharedLock("SimpleCache")

    self.last_cleanup = self._time_fn()

  def _UnlockedCleanup(self):

    """Does cleanup of the cache.

    """

    check_time = self._time_fn()

    if (self.last_cleanup + self.CLEANUP_ROUND) <= check_time:

      keys = []

      for key, value in self.cache.items():

        if not value[TTL]:

          continue

        expired = value[TIMESTAMP] + value[TTL]

        if expired < check_time:

          keys.append(key)

      self._UnlockedInvalidate(keys)

      self.last_cleanup = check_time

  @locking.ssynchronized(_LOCK)

  def Cleanup(self):

    """Cleanup our cache.

    """

    self._UnlockedCleanup()

  @locking.ssynchronized(_LOCK)

  def Store(self, key, value, ttl=0):

    """Stores a value at key in the cache.

    See L{CacheBase.Store} for parameter description

    """

    assert ttl >= 0

    self._UnlockedCleanup()

    val = serializer.Dump(value)

    cache_val = {

      TIMESTAMP: self._time_fn(),

      TTL: ttl,

      VALUE: val

      }

    self.cache[key] = cache_val

    return True

  @locking.ssynchronized(_LOCK, shared=1)

  def GetMulti(self, keys):

    """Retrieve the values of keys from cache.

    See L{CacheBase.GetMulti} for parameter description

    """

    return [self._ExtractValue(key) for key in keys]

  @locking.ssynchronized(_LOCK, shared=1)

  def Get(self, key):

    """Retrieve the value of key from cache.

    See L{CacheBase.Get} for parameter description

    """

    return self._ExtractValue(key)

  @locking.ssynchronized(_LOCK)

  def Invalidate(self, keys):

    """Invalidates value for keys in cache.

    See L{CacheBase.Invalidate} for parameter description

    """

    return self._UnlockedInvalidate(keys)

  @locking.ssynchronized(_LOCK)

  def Flush(self):

    """Invalidates all keys and values in cache.

    See L{CacheBase.Flush} for parameter description

    """

    self.cache.clear()

    self.last_cleanup = self._time_fn()

  def _UnlockedInvalidate(self, keys):

    """Invalidate keys in cache.

    This is the unlocked version, see L{Invalidate} for parameter description

    """

    for key in keys:

      self.cache.pop(key, None)

    return True

  def _ExtractValue(self, key):

    """Extracts just the value for a key.

    This method is taking care if the value did not expire ans returns it

    @param key: The key to look for

    @returns: The value if key is not expired, L{None} otherwise

    """

    try:

      cache_val = self.cache[key]

    except KeyError:

      return None

    else:

      if cache_val[TTL] == 0:

        return serializer.Load(cache_val[VALUE])

      else:

        expired = cache_val[TIMESTAMP] + cache_val[TTL]

        if self._time_fn() <= expired:

          return serializer.Load(cache_val[VALUE])

        else:

          return None

#!/usr/bin/python

#

# Copyright (C) 2011 Google Inc.

#

# This program is free software; you can redistribute it and/or modify

# it under the terms of the GNU General Public License as published by

# the Free Software Foundation; either version 2 of the License, or

# (at your option) any later version.

#

# This program is distributed in the hope that it will be useful, but

# WITHOUT ANY WARRANTY; without even the implied warranty of

# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU

# General Public License for more details.

#

# You should have received a copy of the GNU General Public License

# along with this program; if not, write to the Free Software

# Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA

# 02110-1301, USA.

"""Script for testing ganeti.cache"""

import testutils

import unittest

from ganeti import cache

class ReturnStub:

  def __init__(self, values):

    self.values = values

  def __call__(self):

    assert self.values

    return self.values.pop(0)

class SimpleCacheTest(unittest.TestCase):

  def setUp(self):

    self.cache = cache.SimpleCache()

  def testNoKey(self):

    self.assertEqual(self.cache.GetMulti(["i-dont-exist", "neither-do-i", "no"]),

                     [None, None, None])

  def testCache(self):

    value = 0xc0ffee

    self.assert_(self.cache.Store("i-exist", value))

    self.assertEqual(self.cache.GetMulti(["i-exist"]), [value])

  def testMixed(self):

    value = 0xb4dc0de

    self.assert_(self.cache.Store("i-exist", value))

    self.assertEqual(self.cache.GetMulti(["i-exist", "i-dont"]), [value, None])

  def testTtl(self):

    my_times = ReturnStub([0, 1, 1, 2, 3, 5])

    ttl_cache = cache.SimpleCache(_time_fn=my_times)

    self.assert_(ttl_cache.Store("test-expire", 0xdeadbeef, ttl=2))

    # At this point time will return 2, 1 (start) + 2 (ttl) = 3, still valid

    self.assertEqual(ttl_cache.Get("test-expire"), 0xdeadbeef)

    # At this point time will return 3, 1 (start) + 2 (ttl) = 3, still valid

    self.assertEqual(ttl_cache.Get("test-expire"), 0xdeadbeef)

    # We are at 5, < 3, invalid

    self.assertEqual(ttl_cache.Get("test-expire"), None)

    self.assertFalse(my_times.values)

  def testCleanup(self):

    my_times = ReturnStub([0, 1, 1, 2, 2, 3, 3, 5, 5,

                           21 + cache.SimpleCache.CLEANUP_ROUND,

                           34 + cache.SimpleCache.CLEANUP_ROUND,

                           55 + cache.SimpleCache.CLEANUP_ROUND * 2,

                           89 + cache.SimpleCache.CLEANUP_ROUND * 3])

    # Index 0

    ttl_cache = cache.SimpleCache(_time_fn=my_times)

    # Index 1, 2

    self.assert_(ttl_cache.Store("foobar", 0x1dea, ttl=6))

    # Index 3, 4

    self.assert_(ttl_cache.Store("baz", 0xc0dea55, ttl=11))

    # Index 6, 7

    self.assert_(ttl_cache.Store("long-foobar", "pretty long",

                                 ttl=(22 + cache.SimpleCache.CLEANUP_ROUND)))

    # Index 7, 8

    self.assert_(ttl_cache.Store("foobazbar", "alive forever"))

    self.assertEqual(set(ttl_cache.cache.keys()),

                     set(["foobar", "baz", "long-foobar", "foobazbar"]))

    ttl_cache.Cleanup()

    self.assertEqual(set(ttl_cache.cache.keys()),

                     set(["long-foobar", "foobazbar"]))

    ttl_cache.Cleanup()

    self.assertEqual(set(ttl_cache.cache.keys()),

                     set(["long-foobar", "foobazbar"]))

    ttl_cache.Cleanup()

    self.assertEqual(set(ttl_cache.cache.keys()), set(["foobazbar"]))

    ttl_cache.Cleanup()

    self.assertEqual(set(ttl_cache.cache.keys()), set(["foobazbar"]))

if __name__ == "__main__":

  testutils.GanetiTestProgram()