tahoe-lafs/src/allmydata/storage/crawler.py

489 lines
22 KiB
Python

import os, time, struct
import cPickle as pickle
from twisted.internet import reactor
from twisted.application import service
from allmydata.storage.common import si_b2a
from allmydata.util import fileutil
class TimeSliceExceeded(Exception):
pass
class ShareCrawler(service.MultiService):
"""A ShareCrawler subclass is attached to a StorageServer, and
periodically walks all of its shares, processing each one in some
fashion. This crawl is rate-limited, to reduce the IO burden on the host,
since large servers can easily have a terabyte of shares, in several
million files, which can take hours or days to read.
Once the crawler starts a cycle, it will proceed at a rate limited by the
allowed_cpu_percentage= and cpu_slice= parameters: yielding the reactor
after it has worked for 'cpu_slice' seconds, and not resuming right away,
always trying to use less than 'allowed_cpu_percentage'.
Once the crawler finishes a cycle, it will put off starting the next one
long enough to ensure that 'minimum_cycle_time' elapses between the start
of two consecutive cycles.
We assume that the normal upload/download/get_buckets traffic of a tahoe
grid will cause the prefixdir contents to be mostly cached in the kernel,
or that the number of buckets in each prefixdir will be small enough to
load quickly. A 1TB allmydata.com server was measured to have 2.56M
buckets, spread into the 1024 prefixdirs, with about 2500 buckets per
prefix. On this server, each prefixdir took 130ms-200ms to list the first
time, and 17ms to list the second time.
To use a crawler, create a subclass which implements the process_bucket()
method. It will be called with a prefixdir and a base32 storage index
string. process_bucket() must run synchronously. Any keys added to
self.state will be preserved. Override add_initial_state() to set up
initial state keys. Override finished_cycle() to perform additional
processing when the cycle is complete. Any status that the crawler
produces should be put in the self.state dictionary. Status renderers
(like a web page which describes the accomplishments of your crawler)
will use crawler.get_state() to retrieve this dictionary; they can
present the contents as they see fit.
Then create an instance, with a reference to a StorageServer and a
filename where it can store persistent state. The statefile is used to
keep track of how far around the ring the process has travelled, as well
as timing history to allow the pace to be predicted and controlled. The
statefile will be updated and written to disk after each time slice (just
before the crawler yields to the reactor), and also after each cycle is
finished, and also when stopService() is called. Note that this means
that a crawler which is interrupted with SIGKILL while it is in the
middle of a time slice will lose progress: the next time the node is
started, the crawler will repeat some unknown amount of work.
The crawler instance must be started with startService() before it will
do any work. To make it stop doing work, call stopService().
"""
slow_start = 300 # don't start crawling for 5 minutes after startup
# all three of these can be changed at any time
allowed_cpu_percentage = .10 # use up to 10% of the CPU, on average
cpu_slice = 1.0 # use up to 1.0 seconds before yielding
minimum_cycle_time = 300 # don't run a cycle faster than this
def __init__(self, server, statefile, allowed_cpu_percentage=None):
service.MultiService.__init__(self)
if allowed_cpu_percentage is not None:
self.allowed_cpu_percentage = allowed_cpu_percentage
self.server = server
self.sharedir = server.sharedir
self.statefile = statefile
self.prefixes = [si_b2a(struct.pack(">H", i << (16-10)))[:2]
for i in range(2**10)]
self.prefixes.sort()
self.timer = None
self.bucket_cache = (None, [])
self.current_sleep_time = None
self.next_wake_time = None
self.last_prefix_finished_time = None
self.last_prefix_elapsed_time = None
self.last_cycle_started_time = None
self.last_cycle_elapsed_time = None
self.load_state()
def minus_or_none(self, a, b):
if a is None:
return None
return a-b
def get_progress(self):
"""I return information about how much progress the crawler is
making. My return value is a dictionary. The primary key is
'cycle-in-progress': True if the crawler is currently traversing the
shares, False if it is idle between cycles.
Note that any of these 'time' keys could be None if I am called at
certain moments, so application code must be prepared to tolerate
this case. The estimates will also be None if insufficient data has
been gatherered to form an estimate.
If cycle-in-progress is True, the following keys will be present::
cycle-complete-percentage': float, from 0.0 to 100.0, indicating how
far the crawler has progressed through
the current cycle
remaining-sleep-time: float, seconds from now when we do more work
estimated-cycle-complete-time-left:
float, seconds remaining until the current cycle is finished.
TODO: this does not yet include the remaining time left in
the current prefixdir, and it will be very inaccurate on fast
crawlers (which can process a whole prefix in a single tick)
estimated-time-per-cycle: float, seconds required to do a complete
cycle
If cycle-in-progress is False, the following keys are available::
next-crawl-time: float, seconds-since-epoch when next crawl starts
remaining-wait-time: float, seconds from now when next crawl starts
estimated-time-per-cycle: float, seconds required to do a complete
cycle
"""
d = {}
if self.state["current-cycle"] is None:
d["cycle-in-progress"] = False
d["next-crawl-time"] = self.next_wake_time
d["remaining-wait-time"] = self.minus_or_none(self.next_wake_time,
time.time())
else:
d["cycle-in-progress"] = True
pct = 100.0 * self.last_complete_prefix_index / len(self.prefixes)
d["cycle-complete-percentage"] = pct
remaining = None
if self.last_prefix_elapsed_time is not None:
left = len(self.prefixes) - self.last_complete_prefix_index
remaining = left * self.last_prefix_elapsed_time
# TODO: remainder of this prefix: we need to estimate the
# per-bucket time, probably by measuring the time spent on
# this prefix so far, divided by the number of buckets we've
# processed.
d["estimated-cycle-complete-time-left"] = remaining
# it's possible to call get_progress() from inside a crawler's
# finished_prefix() function
d["remaining-sleep-time"] = self.minus_or_none(self.next_wake_time,
time.time())
per_cycle = None
if self.last_cycle_elapsed_time is not None:
per_cycle = self.last_cycle_elapsed_time
elif self.last_prefix_elapsed_time is not None:
per_cycle = len(self.prefixes) * self.last_prefix_elapsed_time
d["estimated-time-per-cycle"] = per_cycle
return d
def get_state(self):
"""I return the current state of the crawler. This is a copy of my
state dictionary.
If we are not currently sleeping (i.e. get_state() was called from
inside the process_prefixdir, process_bucket, or finished_cycle()
methods, or if startService has not yet been called on this crawler),
these two keys will be None.
Subclasses can override this to add computed keys to the return value,
but don't forget to start with the upcall.
"""
state = self.state.copy() # it isn't a deepcopy, so don't go crazy
return state
def load_state(self):
# we use this to store state for both the crawler's internals and
# anything the subclass-specific code needs. The state is stored
# after each bucket is processed, after each prefixdir is processed,
# and after a cycle is complete. The internal keys we use are:
# ["version"]: int, always 1
# ["last-cycle-finished"]: int, or None if we have not yet finished
# any cycle
# ["current-cycle"]: int, or None if we are sleeping between cycles
# ["current-cycle-start-time"]: int, seconds-since-epoch of when this
# cycle was started, possibly by an earlier
# process
# ["last-complete-prefix"]: str, two-letter name of the last prefixdir
# that was fully processed, or None if we
# are sleeping between cycles, or if we
# have not yet finished any prefixdir since
# a cycle was started
# ["last-complete-bucket"]: str, base32 storage index bucket name
# of the last bucket to be processed, or
# None if we are sleeping between cycles
try:
f = open(self.statefile, "rb")
state = pickle.load(f)
f.close()
except EnvironmentError:
state = {"version": 1,
"last-cycle-finished": None,
"current-cycle": None,
"last-complete-prefix": None,
"last-complete-bucket": None,
}
state.setdefault("current-cycle-start-time", time.time()) # approximate
self.state = state
lcp = state["last-complete-prefix"]
if lcp == None:
self.last_complete_prefix_index = -1
else:
self.last_complete_prefix_index = self.prefixes.index(lcp)
self.add_initial_state()
def add_initial_state(self):
"""Hook method to add extra keys to self.state when first loaded.
The first time this Crawler is used, or when the code has been
upgraded, the saved state file may not contain all the keys you
expect. Use this method to add any missing keys. Simply modify
self.state as needed.
This method for subclasses to override. No upcall is necessary.
"""
pass
def save_state(self):
lcpi = self.last_complete_prefix_index
if lcpi == -1:
last_complete_prefix = None
else:
last_complete_prefix = self.prefixes[lcpi]
self.state["last-complete-prefix"] = last_complete_prefix
tmpfile = self.statefile + ".tmp"
f = open(tmpfile, "wb")
pickle.dump(self.state, f)
f.close()
fileutil.move_into_place(tmpfile, self.statefile)
def startService(self):
# arrange things to look like we were just sleeping, so
# status/progress values work correctly
self.sleeping_between_cycles = True
self.current_sleep_time = self.slow_start
self.next_wake_time = time.time() + self.slow_start
self.timer = reactor.callLater(self.slow_start, self.start_slice)
service.MultiService.startService(self)
def stopService(self):
if self.timer:
self.timer.cancel()
self.timer = None
self.save_state()
return service.MultiService.stopService(self)
def start_slice(self):
start_slice = time.time()
self.timer = None
self.sleeping_between_cycles = False
self.current_sleep_time = None
self.next_wake_time = None
try:
self.start_current_prefix(start_slice)
finished_cycle = True
except TimeSliceExceeded:
finished_cycle = False
self.save_state()
if not self.running:
# someone might have used stopService() to shut us down
return
# either we finished a whole cycle, or we ran out of time
now = time.time()
this_slice = now - start_slice
# this_slice/(this_slice+sleep_time) = percentage
# this_slice/percentage = this_slice+sleep_time
# sleep_time = (this_slice/percentage) - this_slice
sleep_time = (this_slice / self.allowed_cpu_percentage) - this_slice
# if the math gets weird, or a timequake happens, don't sleep
# forever. Note that this means that, while a cycle is running, we
# will process at least one bucket every 5 minutes, no matter how
# long that bucket takes.
sleep_time = max(0.0, min(sleep_time, 299))
if finished_cycle:
# how long should we sleep between cycles? Don't run faster than
# allowed_cpu_percentage says, but also run faster than
# minimum_cycle_time
self.sleeping_between_cycles = True
sleep_time = max(sleep_time, self.minimum_cycle_time)
else:
self.sleeping_between_cycles = False
self.current_sleep_time = sleep_time # for status page
self.next_wake_time = now + sleep_time
self.yielding(sleep_time)
self.timer = reactor.callLater(sleep_time, self.start_slice)
def start_current_prefix(self, start_slice):
state = self.state
if state["current-cycle"] is None:
self.last_cycle_started_time = time.time()
state["current-cycle-start-time"] = self.last_cycle_started_time
if state["last-cycle-finished"] is None:
state["current-cycle"] = 0
else:
state["current-cycle"] = state["last-cycle-finished"] + 1
self.started_cycle(state["current-cycle"])
cycle = state["current-cycle"]
for i in range(self.last_complete_prefix_index+1, len(self.prefixes)):
# if we want to yield earlier, just raise TimeSliceExceeded()
prefix = self.prefixes[i]
prefixdir = os.path.join(self.sharedir, prefix)
if i == self.bucket_cache[0]:
buckets = self.bucket_cache[1]
else:
try:
buckets = os.listdir(prefixdir)
buckets.sort()
except EnvironmentError:
buckets = []
self.bucket_cache = (i, buckets)
self.process_prefixdir(cycle, prefix, prefixdir,
buckets, start_slice)
self.last_complete_prefix_index = i
now = time.time()
if self.last_prefix_finished_time is not None:
elapsed = now - self.last_prefix_finished_time
self.last_prefix_elapsed_time = elapsed
self.last_prefix_finished_time = now
self.finished_prefix(cycle, prefix)
if time.time() >= start_slice + self.cpu_slice:
raise TimeSliceExceeded()
# yay! we finished the whole cycle
self.last_complete_prefix_index = -1
self.last_prefix_finished_time = None # don't include the sleep
now = time.time()
if self.last_cycle_started_time is not None:
self.last_cycle_elapsed_time = now - self.last_cycle_started_time
state["last-complete-bucket"] = None
state["last-cycle-finished"] = cycle
state["current-cycle"] = None
self.finished_cycle(cycle)
self.save_state()
def process_prefixdir(self, cycle, prefix, prefixdir, buckets, start_slice):
"""This gets a list of bucket names (i.e. storage index strings,
base32-encoded) in sorted order.
You can override this if your crawler doesn't care about the actual
shares, for example a crawler which merely keeps track of how many
buckets are being managed by this server.
Subclasses which *do* care about actual bucket should leave this
method along, and implement process_bucket() instead.
"""
for bucket in buckets:
if bucket <= self.state["last-complete-bucket"]:
continue
self.process_bucket(cycle, prefix, prefixdir, bucket)
self.state["last-complete-bucket"] = bucket
if time.time() >= start_slice + self.cpu_slice:
raise TimeSliceExceeded()
# the remaining methods are explictly for subclasses to implement.
def started_cycle(self, cycle):
"""Notify a subclass that the crawler is about to start a cycle.
This method is for subclasses to override. No upcall is necessary.
"""
pass
def process_bucket(self, cycle, prefix, prefixdir, storage_index_b32):
"""Examine a single bucket. Subclasses should do whatever they want
to do to the shares therein, then update self.state as necessary.
If the crawler is never interrupted by SIGKILL, this method will be
called exactly once per share (per cycle). If it *is* interrupted,
then the next time the node is started, some amount of work will be
duplicated, according to when self.save_state() was last called. By
default, save_state() is called at the end of each timeslice, and
after finished_cycle() returns, and when stopService() is called.
To reduce the chance of duplicate work (i.e. to avoid adding multiple
records to a database), you can call save_state() at the end of your
process_bucket() method. This will reduce the maximum duplicated work
to one bucket per SIGKILL. It will also add overhead, probably 1-20ms
per bucket (and some disk writes), which will count against your
allowed_cpu_percentage, and which may be considerable if
process_bucket() runs quickly.
This method is for subclasses to override. No upcall is necessary.
"""
pass
def finished_prefix(self, cycle, prefix):
"""Notify a subclass that the crawler has just finished processing a
prefix directory (all buckets with the same two-character/10bit
prefix). To impose a limit on how much work might be duplicated by a
SIGKILL that occurs during a timeslice, you can call
self.save_state() here, but be aware that it may represent a
significant performance hit.
This method is for subclasses to override. No upcall is necessary.
"""
pass
def finished_cycle(self, cycle):
"""Notify subclass that a cycle (one complete traversal of all
prefixdirs) has just finished. 'cycle' is the number of the cycle
that just finished. This method should perform summary work and
update self.state to publish information to status displays.
One-shot crawlers, such as those used to upgrade shares to a new
format or populate a database for the first time, can call
self.stopService() (or more likely self.disownServiceParent()) to
prevent it from running a second time. Don't forget to set some
persistent state so that the upgrader won't be run again the next
time the node is started.
This method is for subclasses to override. No upcall is necessary.
"""
pass
def yielding(self, sleep_time):
"""The crawler is about to sleep for 'sleep_time' seconds. This
method is mostly for the convenience of unit tests.
This method is for subclasses to override. No upcall is necessary.
"""
pass
class BucketCountingCrawler(ShareCrawler):
"""I keep track of how many buckets are being managed by this server.
This is equivalent to the number of distributed files and directories for
which I am providing storage. The actual number of files+directories in
the full grid is probably higher (especially when there are more servers
than 'N', the number of generated shares), because some files+directories
will have shares on other servers instead of me. Also note that the
number of buckets will differ from the number of shares in small grids,
when more than one share is placed on a single server.
"""
minimum_cycle_time = 60*60 # we don't need this more than once an hour
def __init__(self, server, statefile, num_sample_prefixes=1):
ShareCrawler.__init__(self, server, statefile)
self.num_sample_prefixes = num_sample_prefixes
def add_initial_state(self):
# ["bucket-counts"][cyclenum][prefix] = number
# ["last-complete-cycle"] = cyclenum # maintained by base class
# ["last-complete-bucket-count"] = number
# ["storage-index-samples"][prefix] = (cyclenum,
# list of SI strings (base32))
self.state.setdefault("bucket-counts", {})
self.state.setdefault("last-complete-bucket-count", None)
self.state.setdefault("storage-index-samples", {})
def process_prefixdir(self, cycle, prefix, prefixdir, buckets, start_slice):
# we override process_prefixdir() because we don't want to look at
# the individual buckets. We'll save state after each one. On my
# laptop, a mostly-empty storage server can process about 70
# prefixdirs in a 1.0s slice.
if cycle not in self.state["bucket-counts"]:
self.state["bucket-counts"][cycle] = {}
self.state["bucket-counts"][cycle][prefix] = len(buckets)
if prefix in self.prefixes[:self.num_sample_prefixes]:
self.state["storage-index-samples"][prefix] = (cycle, buckets)
def finished_cycle(self, cycle):
last_counts = self.state["bucket-counts"].get(cycle, [])
if len(last_counts) == len(self.prefixes):
# great, we have a whole cycle.
num_buckets = sum(last_counts.values())
self.state["last-complete-bucket-count"] = num_buckets
# get rid of old counts
for old_cycle in list(self.state["bucket-counts"].keys()):
if old_cycle != cycle:
del self.state["bucket-counts"][old_cycle]
# get rid of old samples too
for prefix in list(self.state["storage-index-samples"].keys()):
old_cycle,buckets = self.state["storage-index-samples"][prefix]
if old_cycle != cycle:
del self.state["storage-index-samples"][prefix]