# # Copyright (c) 2011-2014 Red Hat, Inc. # This file is part of GlusterFS. # This file is licensed to you under your choice of the GNU Lesser # General Public License, version 3 or any later version (LGPLv3 or # later), or the GNU General Public License, version 2 (GPLv2), in all # cases as published by the Free Software Foundation. # import os import sys import time import stat import json import logging import fcntl import string import errno import tarfile from errno import ENOENT, ENODATA, EEXIST, EACCES, EAGAIN, ESTALE from threading import Condition, Lock from datetime import datetime from gconf import gconf from syncdutils import Thread, GsyncdError, boolify, escape from syncdutils import unescape, gauxpfx, md5hex, selfkill from syncdutils import lstat, errno_wrap, FreeObject from syncdutils import NoPurgeTimeAvailable, PartialHistoryAvailable URXTIME = (-1, 0) # Default rollover time set in changelog translator # changelog rollover time is hardcoded here to avoid the # xsync usage when crawling switch happens from history # to changelog. If rollover time increased in translator # then geo-rep can enter into xsync crawl after history # crawl before starting live changelog crawl. CHANGELOG_ROLLOVER_TIME = 15 # Max size of Changelogs to process per batch, Changelogs Processing is # not limited by the number of changelogs but instead based on # size of the changelog file, One sample changelog file size was 145408 # with ~1000 CREATE and ~1000 DATA. 5 such files in one batch is 727040 # If geo-rep worker crashes while processing a batch, it has to retry only # that batch since stime will get updated after each batch. MAX_CHANGELOG_BATCH_SIZE = 727040 # Utility functions to help us to get to closer proximity # of the DRY principle (no, don't look for elevated or # perspectivistic things here) def _xtime_now(): t = time.time() sec = int(t) nsec = int((t - sec) * 1000000) return (sec, nsec) def _volinfo_hook_relax_foreign(self): volinfo_sys = self.get_sys_volinfo() fgn_vi = volinfo_sys[self.KFGN] if fgn_vi: expiry = fgn_vi['timeout'] - int(time.time()) + 1 logging.info('foreign volume info found, waiting %d sec for expiry' % expiry) time.sleep(expiry) volinfo_sys = self.get_sys_volinfo() return volinfo_sys # The API! def gmaster_builder(excrawl=None): """produce the GMaster class variant corresponding to sync mode""" this = sys.modules[__name__] modemixin = gconf.special_sync_mode if not modemixin: modemixin = 'normal' changemixin = 'xsync' if gconf.change_detector == 'xsync' \ else excrawl or gconf.change_detector logging.info('setting up %s change detection mode' % changemixin) modemixin = getattr(this, modemixin.capitalize() + 'Mixin') crawlmixin = getattr(this, 'GMaster' + changemixin.capitalize() + 'Mixin') sendmarkmixin = boolify( gconf.use_rsync_xattrs) and SendmarkRsyncMixin or SendmarkNormalMixin purgemixin = boolify( gconf.ignore_deletes) and PurgeNoopMixin or PurgeNormalMixin syncengine = boolify(gconf.use_tarssh) and TarSSHEngine or RsyncEngine class _GMaster(crawlmixin, modemixin, sendmarkmixin, purgemixin, syncengine): pass return _GMaster # Mixin classes that implement the data format # and logic particularities of the certain # sync modes class NormalMixin(object): """normal geo-rep behavior""" minus_infinity = URXTIME # following staticmethods ideally would be # methods of an xtime object (in particular, # implementing the hooks needed for comparison # operators), but at this point we don't yet # have a dedicated xtime class @staticmethod def serialize_xtime(xt): return "%d.%d" % tuple(xt) @staticmethod def deserialize_xtime(xt): return tuple(int(x) for x in xt.split(".")) @staticmethod def native_xtime(xt): return xt @staticmethod def xtime_geq(xt0, xt1): return xt0 >= xt1 def make_xtime_opts(self, is_master, opts): if not 'create' in opts: opts['create'] = is_master if not 'default_xtime' in opts: opts['default_xtime'] = URXTIME def xtime_low(self, rsc, path, **opts): if rsc == self.master: xt = rsc.server.xtime(path, self.uuid) else: xt = rsc.server.stime(path, self.uuid) if isinstance(xt, int) and xt == ENODATA: xt = rsc.server.xtime(path, self.uuid) if not isinstance(xt, int): self.slave.server.set_stime(path, self.uuid, xt) if isinstance(xt, int) and xt != ENODATA: return xt if xt == ENODATA or xt < self.volmark: if opts['create']: xt = _xtime_now() rsc.server.aggregated.set_xtime(path, self.uuid, xt) else: xt = opts['default_xtime'] return xt def keepalive_payload_hook(self, timo, gap): # first grab a reference as self.volinfo # can be changed in main thread vi = self.volinfo if vi: # then have a private copy which we can mod vi = vi.copy() vi['timeout'] = int(time.time()) + timo else: # send keep-alives more frequently to # avoid a delay in announcing our volume info # to slave if it becomes established in the # meantime gap = min(10, gap) return (vi, gap) def volinfo_hook(self): return self.get_sys_volinfo() def xtime_reversion_hook(self, path, xtl, xtr): if xtr > xtl: raise GsyncdError("timestamp corruption for " + path) def need_sync(self, e, xte, xtrd): return xte > xtrd def set_slave_xtime(self, path, mark): self.slave.server.set_stime(path, self.uuid, mark) # self.slave.server.set_xtime_remote(path, self.uuid, mark) class PartialMixin(NormalMixin): """a variant tuned towards operation with a master that has partial info of the slave (brick typically)""" def xtime_reversion_hook(self, path, xtl, xtr): pass class RecoverMixin(NormalMixin): """a variant that differs from normal in terms of ignoring non-indexed files""" @staticmethod def make_xtime_opts(is_master, opts): if not 'create' in opts: opts['create'] = False if not 'default_xtime' in opts: opts['default_xtime'] = URXTIME def keepalive_payload_hook(self, timo, gap): return (None, gap) def volinfo_hook(self): return _volinfo_hook_relax_foreign(self) # Further mixins for certain tunable behaviors class SendmarkNormalMixin(object): def sendmark_regular(self, *a, **kw): return self.sendmark(*a, **kw) class SendmarkRsyncMixin(object): def sendmark_regular(self, *a, **kw): pass class PurgeNormalMixin(object): def purge_missing(self, path, names): self.slave.server.purge(path, names) class PurgeNoopMixin(object): def purge_missing(self, path, names): pass class TarSSHEngine(object): """Sync engine that uses tar(1) piped over ssh(1) for data transfers. Good for lots of small files. """ def a_syncdata(self, files): logging.debug('files: %s' % (files)) for f in files: pb = self.syncer.add(f) def regjob(se, xte, pb): rv = pb.wait() if rv[0]: logging.debug('synced ' + se) return True else: # stat check for file presence st = lstat(se) if isinstance(st, int): # file got unlinked in the interim self.unlinked_gfids.add(se) return True self.add_job(self.FLAT_DIR_HIERARCHY, 'reg', regjob, f, None, pb) def syncdata_wait(self): if self.wait(self.FLAT_DIR_HIERARCHY, None): return True def syncdata(self, files): self.a_syncdata(files) self.syncdata_wait() class RsyncEngine(object): """Sync engine that uses rsync(1) for data transfers""" def a_syncdata(self, files): logging.debug('files: %s' % (files)) for f in files: logging.debug('candidate for syncing %s' % f) pb = self.syncer.add(f) def regjob(se, xte, pb): rv = pb.wait() if rv[0]: logging.debug('synced ' + se) return True else: # stat to check if the file exist st = lstat(se) if isinstance(st, int): # file got unlinked in the interim self.unlinked_gfids.add(se) return True self.add_job(self.FLAT_DIR_HIERARCHY, 'reg', regjob, f, None, pb) def syncdata_wait(self): if self.wait(self.FLAT_DIR_HIERARCHY, None): return True def syncdata(self, files): self.a_syncdata(files) self.syncdata_wait() class GMasterCommon(object): """abstract class impementling master role""" KFGN = 0 KNAT = 1 def get_sys_volinfo(self): """query volume marks on fs root err out on multiple foreign masters """ fgn_vis, nat_vi = ( self.master.server.aggregated.foreign_volume_infos(), self.master.server.aggregated.native_volume_info()) fgn_vi = None if fgn_vis: if len(fgn_vis) > 1: raise GsyncdError("cannot work with multiple foreign masters") fgn_vi = fgn_vis[0] return fgn_vi, nat_vi @property def uuid(self): if self.volinfo: return self.volinfo['uuid'] @property def volmark(self): if self.volinfo: return self.volinfo['volume_mark'] def xtime(self, path, *a, **opts): """get amended xtime as of amending, we can create missing xtime, or determine a valid value if what we get is expired (as of the volume mark expiry); way of amendig depends on @opts and on subject of query (master or slave). """ if a: rsc = a[0] else: rsc = self.master self.make_xtime_opts(rsc == self.master, opts) return self.xtime_low(rsc, path, **opts) def __init__(self, master, slave): self.master = master self.slave = slave self.jobtab = {} if boolify(gconf.use_tarssh): logging.info("using 'tar over ssh' as the sync engine") self.syncer = Syncer(slave, self.slave.tarssh, [2]) else: logging.info("using 'rsync' as the sync engine") # partial transfer (cf. rsync(1)), that's normal self.syncer = Syncer(slave, self.slave.rsync, [23, 24]) # crawls vs. turns: # - self.crawls is simply the number of crawl() invocations on root # - one turn is a maximal consecutive sequence of crawls so that each # crawl in it detects a change to be synced # - self.turns is the number of turns since start # - self.total_turns is a limit so that if self.turns reaches it, then # we exit (for diagnostic purposes) # so, eg., if the master fs changes unceasingly, self.turns will remain # 0. self.crawls = 0 self.turns = 0 self.total_turns = int(gconf.turns) self.crawl_start = datetime.now() self.lastreport = {'crawls': 0, 'turns': 0, 'time': 0} self.start = None self.change_seen = None # the actual volinfo we make use of self.volinfo = None self.terminate = False self.sleep_interval = 1 self.unlinked_gfids = set() def init_keep_alive(cls): """start the keep-alive thread """ timo = int(gconf.timeout or 0) if timo > 0: def keep_alive(): while True: vi, gap = cls.keepalive_payload_hook(timo, timo * 0.5) cls.slave.server.keep_alive(vi) time.sleep(gap) t = Thread(target=keep_alive) t.start() def mgmt_lock(self): """Take management volume lock """ if gconf.mgmt_lock_fd: try: fcntl.lockf(gconf.mgmt_lock_fd, fcntl.LOCK_EX | fcntl.LOCK_NB) if not gconf.active_earlier: gconf.active_earlier = True logging.info("Got lock : %s : Becoming ACTIVE" % gconf.local_path) return True except: ex = sys.exc_info()[1] if isinstance(ex, IOError) and ex.errno in (EACCES, EAGAIN): if not gconf.passive_earlier: gconf.passive_earlier = True logging.info("Didn't get lock : %s : Becoming PASSIVE" % gconf.local_path) return False raise fd = None bname = str(self.uuid) + "_" + str(gconf.slave_id) + "_subvol_" \ + str(gconf.subvol_num) + ".lock" mgmt_lock_dir = os.path.join(gconf.meta_volume_mnt, "geo-rep") path = os.path.join(mgmt_lock_dir, bname) logging.debug("lock_file_path: %s" % path) try: fd = os.open(path, os.O_CREAT | os.O_RDWR) except OSError: ex = sys.exc_info()[1] if ex.errno == ENOENT: logging.info("Creating geo-rep directory in meta volume...") try: os.makedirs(mgmt_lock_dir) except OSError: ex = sys.exc_info()[1] if ex.errno == EEXIST: pass else: raise fd = os.open(path, os.O_CREAT | os.O_RDWR) else: raise try: fcntl.lockf(fd, fcntl.LOCK_EX | fcntl.LOCK_NB) # Save latest FD for future use gconf.mgmt_lock_fd = fd except: ex = sys.exc_info()[1] if isinstance(ex, IOError) and ex.errno in (EACCES, EAGAIN): # cannot grab, it's taken if not gconf.passive_earlier: gconf.passive_earlier = True logging.info("Didn't get lock : %s : Becoming PASSIVE" % gconf.local_path) gconf.mgmt_lock_fd = fd return False raise if not gconf.active_earlier: gconf.active_earlier = True logging.info("Got lock : %s : Becoming ACTIVE" % gconf.local_path) return True def should_crawl(self): if not boolify(gconf.use_meta_volume): return gconf.glusterd_uuid in self.master.server.node_uuid() if not os.path.ismount(gconf.meta_volume_mnt): logging.error("Meta-volume is not mounted. Worker Exiting...") sys.exit(1) return self.mgmt_lock() def register(self): self.register() def crawlwrap(self, oneshot=False, register_time=None): if oneshot: # it's important to do this during the oneshot crawl as # for a passive gsyncd (ie. in a replicate scenario) # the keepalive thread would keep the connection alive. self.init_keep_alive() # If crawlwrap is called when partial history available, # then it sets register_time which is the time when geo-rep # worker registerd to changelog consumption. Since nsec is # not considered in register time, their are chances of skipping # changes detection in xsync crawl. This limit will be reset when # crawlwrap is called again. self.live_changelog_start_time = None if register_time: self.live_changelog_start_time = (register_time, 0) # no need to maintain volinfo state machine. # in a cascading setup, each geo-replication session is # independent (ie. 'volume-mark' and 'xtime' are not # propogated). This is because the slave's xtime is now # stored on the master itself. 'volume-mark' just identifies # that we are in a cascading setup and need to enable # 'geo-replication.ignore-pid-check' option. volinfo_sys = self.volinfo_hook() self.volinfo = volinfo_sys[self.KNAT] inter_master = volinfo_sys[self.KFGN] logging.info("%s master with volume id %s ..." % (inter_master and "intermediate" or "primary", self.uuid)) gconf.configinterface.set('volume_id', self.uuid) if self.volinfo: if self.volinfo['retval']: logging.warn("master cluster's info may not be valid %d" % self.volinfo['retval']) else: raise GsyncdError("master volinfo unavailable") self.lastreport['time'] = time.time() logging.info('crawl interval: %d seconds' % self.sleep_interval) t0 = time.time() crawl = self.should_crawl() while not self.terminate: if self.start: logging.debug("... crawl #%d done, took %.6f seconds" % (self.crawls, time.time() - self.start)) self.start = time.time() should_display_info = self.start - self.lastreport['time'] >= 60 if should_display_info: logging.info("%d crawls, %d turns", self.crawls - self.lastreport['crawls'], self.turns - self.lastreport['turns']) self.lastreport.update(crawls=self.crawls, turns=self.turns, time=self.start) t1 = time.time() if int(t1 - t0) >= int(gconf.replica_failover_interval): crawl = self.should_crawl() t0 = t1 self.update_worker_remote_node() if not crawl: self.status.set_passive() # bring up _this_ brick to the cluster stime # which is min of cluster (but max of the replicas) brick_stime = self.xtime('.', self.slave) cluster_stime = self.master.server.aggregated.stime_mnt( '.', '.'.join([str(self.uuid), str(gconf.slave_id)])) logging.debug("Cluster stime: %s | Brick stime: %s" % (repr(cluster_stime), repr(brick_stime))) if not isinstance(cluster_stime, int): if brick_stime < cluster_stime: self.slave.server.set_stime( self.FLAT_DIR_HIERARCHY, self.uuid, cluster_stime) # Purge all changelogs available in processing dir # less than cluster_stime proc_dir = os.path.join(self.setup_working_dir(), ".processing") if os.path.exists(proc_dir): to_purge = [f for f in os.listdir(proc_dir) if (f.startswith("CHANGELOG.") and int(f.split('.')[-1]) < cluster_stime[0])] for f in to_purge: os.remove(os.path.join(proc_dir, f)) time.sleep(5) continue self.status.set_active() self.crawl() if oneshot: return time.sleep(self.sleep_interval) @staticmethod def humantime(*tpair): """format xtime-like (sec, nsec) pair to human readable format""" ts = datetime.fromtimestamp(float('.'.join(str(n) for n in tpair))).\ strftime("%Y-%m-%d %H:%M:%S") if len(tpair) > 1: ts += '.' + str(tpair[1]) return ts def _crawl_time_format(self, crawl_time): # Ex: 5 years, 4 days, 20:23:10 years, days = divmod(crawl_time.days, 365.25) years = int(years) days = int(days) date = "" m, s = divmod(crawl_time.seconds, 60) h, m = divmod(m, 60) if years != 0: date += "%s %s " % (years, "year" if years == 1 else "years") if days != 0: date += "%s %s " % (days, "day" if days == 1 else "days") date += "%s:%s:%s" % (string.zfill(h, 2), string.zfill(m, 2), string.zfill(s, 2)) return date def add_job(self, path, label, job, *a, **kw): """insert @job function to job table at @path with @label""" if self.jobtab.get(path) is None: self.jobtab[path] = [] self.jobtab[path].append((label, a, lambda: job(*a, **kw))) def add_failjob(self, path, label): """invoke .add_job with a job that does nothing just fails""" logging.debug('salvaged: ' + label) self.add_job(path, label, lambda: False) def wait(self, path, *args): """perform jobs registered for @path Reset jobtab entry for @path, determine success as the conjunction of success of all the jobs. In case of success, call .sendmark on @path """ jobs = self.jobtab.pop(path, []) succeed = True for j in jobs: ret = j[-1]() if not ret: succeed = False if succeed and not args[0] is None: self.sendmark(path, *args) return succeed def sendmark(self, path, mark, adct=None): """update slave side xtime for @path to master side xtime also can send a setattr payload (see Server.setattr). """ if adct: self.slave.server.setattr(path, adct) self.set_slave_xtime(path, mark) class XCrawlMetadata(object): def __init__(self, st_uid, st_gid, st_mode, st_atime, st_mtime): self.st_uid = int(st_uid) self.st_gid = int(st_gid) self.st_mode = int(st_mode) self.st_atime = float(st_atime) self.st_mtime = float(st_mtime) class GMasterChangelogMixin(GMasterCommon): """ changelog based change detection and syncing """ # index for change type and entry IDX_START = 0 IDX_END = 2 UNLINK_ENTRY = 2 POS_GFID = 0 POS_TYPE = 1 POS_ENTRY1 = -1 TYPE_META = "M " TYPE_GFID = "D " TYPE_ENTRY = "E " # flat directory hierarchy for gfid based access FLAT_DIR_HIERARCHY = '.' # maximum retries per changelog before giving up MAX_RETRIES = 10 CHANGELOG_LOG_LEVEL = 9 CHANGELOG_CONN_RETRIES = 5 def archive_and_purge_changelogs(self, changelogs): # Creates tar file instead of tar.gz, since changelogs will # be appended to existing tar. archive name is # archive_.tar archive_name = "archive_%s.tar" % datetime.today().strftime( gconf.changelog_archive_format) try: tar = tarfile.open(os.path.join(self.processed_changelogs_dir, archive_name), "a") except tarfile.ReadError: tar = tarfile.open(os.path.join(self.processed_changelogs_dir, archive_name), "w") for f in changelogs: try: f = os.path.basename(f) tar.add(os.path.join(self.processed_changelogs_dir, f), arcname=os.path.basename(f)) except: exc = sys.exc_info()[1] if ((isinstance(exc, OSError) or isinstance(exc, IOError)) and exc.errno == ENOENT): continue else: tar.close() raise tar.close() for f in changelogs: try: f = os.path.basename(f) os.remove(os.path.join(self.processed_changelogs_dir, f)) except OSError as e: if e.errno == errno.ENOENT: continue else: raise def fallback_xsync(self): logging.info('falling back to xsync mode') gconf.configinterface.set('change-detector', 'xsync') selfkill() def setup_working_dir(self): workdir = os.path.join(gconf.working_dir, md5hex(gconf.local_path)) logging.debug('changelog working dir %s' % workdir) return workdir def get_purge_time(self): purge_time = self.xtime('.', self.slave) if isinstance(purge_time, int): purge_time = None return purge_time def process_change(self, change, done, retry): pfx = gauxpfx() clist = [] entries = [] meta_gfid = set() datas = set() # basic crawl stats: files and bytes files_pending = {'count': 0, 'purge': 0, 'bytes': 0, 'files': []} try: f = open(change, "r") clist = f.readlines() f.close() except IOError: raise def edct(op, **ed): dct = {} dct['op'] = op for k in ed: if k == 'stat': st = ed[k] dst = dct['stat'] = {} if st: dst['uid'] = st.st_uid dst['gid'] = st.st_gid dst['mode'] = st.st_mode dst['atime'] = st.st_atime dst['mtime'] = st.st_mtime else: dct[k] = ed[k] return dct # entry counts (not purges) def entry_update(): files_pending['count'] += 1 # purge count def purge_update(): files_pending['purge'] += 1 def log_failures(failures, entry_key, gfid_prefix, log_prefix): num_failures = 0 for failure in failures: st = lstat(os.path.join(gfid_prefix, failure[0][entry_key])) if not isinstance(st, int): num_failures += 1 logging.error('%s FAILED: %s' % (log_prefix, repr(failure))) self.status.inc_value("failures", num_failures) for e in clist: e = e.strip() et = e[self.IDX_START:self.IDX_END] # entry type ec = e[self.IDX_END:].split(' ') # rest of the bits # skip ENTRY operation if hot tier brick if self.name == 'live_changelog' or \ self.name == 'history_changelog': if boolify(gconf.is_hottier) and et == self.TYPE_ENTRY: logging.debug('skip ENTRY op: %s if hot tier brick' % (ec[self.POS_TYPE])) continue if et == self.TYPE_ENTRY: # extract information according to the type of # the entry operation. create(), mkdir() and mknod() # have mode, uid, gid information in the changelog # itself, so no need to stat()... ty = ec[self.POS_TYPE] # PARGFID/BNAME en = unescape(os.path.join(pfx, ec[self.POS_ENTRY1])) # GFID of the entry gfid = ec[self.POS_GFID] if ty in ['UNLINK', 'RMDIR']: # The index of PARGFID/BNAME for UNLINK, RMDIR # is no more the last index. It varies based on # changelog.capture-del-path is enabled or not. en = unescape(os.path.join(pfx, ec[self.UNLINK_ENTRY])) # Remove from DATA list, so that rsync will # not fail pt = os.path.join(pfx, ec[0]) if pt in datas: datas.remove(pt) if not boolify(gconf.ignore_deletes): purge_update() entries.append(edct(ty, gfid=gfid, entry=en)) elif ty in ['CREATE', 'MKDIR', 'MKNOD']: entry_update() # Special case: record mknod as link if ty in ['MKNOD']: mode = int(ec[2]) if mode & 01000: # Avoid stat'ing the file as it # may be deleted in the interim st = FreeObject(st_mode=int(ec[2]), st_uid=int(ec[3]), st_gid=int(ec[4]), st_atime=0, st_mtime=0) # So, it may be deleted, but still we are # append LINK? Because, the file will be # CREATED if source not exists. entries.append(edct('LINK', stat=st, entry=en, gfid=gfid)) # Here, we have the assumption that only # tier-gfid.linkto causes this mknod. Add data datas.add(os.path.join(pfx, ec[0])) continue # stat info. present in the changelog itself entries.append(edct(ty, gfid=gfid, entry=en, mode=int(ec[2]), uid=int(ec[3]), gid=int(ec[4]))) elif ty == "RENAME": go = os.path.join(pfx, gfid) st = lstat(go) if isinstance(st, int): st = {} rl = None if st and stat.S_ISLNK(st.st_mode): rl = errno_wrap(os.readlink, [en], [ENOENT], [ESTALE]) if isinstance(rl, int): rl = None entry_update() e1 = unescape(os.path.join(pfx, ec[self.POS_ENTRY1 - 1])) entries.append(edct(ty, gfid=gfid, entry=e1, entry1=en, stat=st, link=rl)) else: # stat() to get mode and other information go = os.path.join(pfx, gfid) st = lstat(go) if isinstance(st, int): logging.debug('file %s got purged in the interim' % go) continue if ty == 'LINK': entry_update() entries.append(edct(ty, stat=st, entry=en, gfid=gfid)) elif ty == 'SYMLINK': rl = errno_wrap(os.readlink, [en], [ENOENT], [ESTALE]) if isinstance(rl, int): continue entry_update() entries.append( edct(ty, stat=st, entry=en, gfid=gfid, link=rl)) else: logging.warn('ignoring %s [op %s]' % (gfid, ty)) elif et == self.TYPE_GFID: # If self.unlinked_gfids is available, then that means it is # retrying the changelog second time. Do not add the GFID's # to rsync job if failed previously but unlinked in master if self.unlinked_gfids and \ os.path.join(pfx, ec[0]) in self.unlinked_gfids: logging.debug("ignoring data, since file purged interim") else: datas.add(os.path.join(pfx, ec[0])) elif et == self.TYPE_META: if ec[1] == 'SETATTR': # only setattr's for now... if len(ec) == 5: # In xsync crawl, we already have stat data # avoid doing stat again meta_gfid.add((os.path.join(pfx, ec[0]), XCrawlMetadata(st_uid=ec[2], st_gid=ec[3], st_mode=ec[4], st_atime=ec[5], st_mtime=ec[6]))) else: meta_gfid.add((os.path.join(pfx, ec[0]), )) elif ec[1] == 'SETXATTR' or ec[1] == 'XATTROP' or \ ec[1] == 'FXATTROP': # To sync xattr/acls use rsync/tar, --xattrs and --acls # switch to rsync and tar if not boolify(gconf.use_tarssh) and \ (boolify(gconf.sync_xattrs) or boolify(gconf.sync_acls)): datas.add(os.path.join(pfx, ec[0])) else: logging.warn('got invalid changelog type: %s' % (et)) logging.debug('entries: %s' % repr(entries)) # Increment counters for Status self.status.inc_value("entry", len(entries)) self.files_in_batch += len(datas) self.status.inc_value("data", len(datas)) # sync namespace if entries: failures = self.slave.server.entry_ops(entries) log_failures(failures, 'gfid', gauxpfx(), 'ENTRY') self.status.dec_value("entry", len(entries)) # sync metadata if meta_gfid: meta_entries = [] for go in meta_gfid: if len(go) > 1: st = go[1] else: st = lstat(go[0]) if isinstance(st, int): logging.debug('file %s got purged in the interim' % go[0]) continue meta_entries.append(edct('META', go=go[0], stat=st)) if meta_entries: self.status.inc_value("meta", len(entries)) failures = self.slave.server.meta_ops(meta_entries) log_failures(failures, 'go', '', 'META') self.status.dec_value("meta", len(entries)) # sync data if datas: self.a_syncdata(datas) self.datas_in_batch.update(datas) def process(self, changes, done=1): tries = 0 retry = False self.unlinked_gfids = set() self.files_in_batch = 0 self.datas_in_batch = set() self.syncer.disable_errorlog() while True: # first, fire all changelog transfers in parallel. entry and # metadata are performed synchronously, therefore in serial. # However at the end of each changelog, data is synchronized # with syncdata_async() - which means it is serial w.r.t # entries/metadata of that changelog but happens in parallel # with data of other changelogs. if retry: if tries == (self.MAX_RETRIES - 1): # Enable Error logging if it is last retry self.syncer.enable_errorlog() # Remove Unlinked GFIDs from Queue for unlinked_gfid in self.unlinked_gfids: self.datas_in_batch.remove(unlinked_gfid) # Retry only Sync. Do not retry entry ops if self.datas_in_batch: self.a_syncdata(self.datas_in_batch) else: for change in changes: logging.debug('processing change %s' % change) self.process_change(change, done, retry) if not retry: # number of changelogs processed in the batch self.turns += 1 # Now we wait for all the data transfers fired off in the above # step to complete. Note that this is not ideal either. Ideally # we want to trigger the entry/meta-data transfer of the next # batch while waiting for the data transfer of the current batch # to finish. # Note that the reason to wait for the data transfer (vs doing it # completely in the background and call the changelog_done() # asynchronously) is because this waiting acts as a "backpressure" # and prevents a spiraling increase of wait stubs from consuming # unbounded memory and resources. # update the slave's time with the timestamp of the _last_ # changelog file time suffix. Since, the changelog prefix time # is the time when the changelog was rolled over, introduce a # tolerance of 1 second to counter the small delta b/w the # marker update and gettimeofday(). # NOTE: this is only for changelog mode, not xsync. # @change is the last changelog (therefore max time for this batch) if self.syncdata_wait(): self.unlinked_gfids = set() if done: xtl = (int(change.split('.')[-1]) - 1, 0) self.upd_stime(xtl) map(self.changelog_done_func, changes) self.archive_and_purge_changelogs(changes) # Reset Data counter after sync self.status.dec_value("data", self.files_in_batch) self.files_in_batch = 0 self.datas_in_batch = set() break # We do not know which changelog transfer failed, retry everything. retry = True tries += 1 if tries == self.MAX_RETRIES: logging.error('changelogs %s could not be processed ' 'completely - moving on...' % ' '.join(map(os.path.basename, changes))) # Reset data counter on failure self.status.dec_value("data", self.files_in_batch) self.files_in_batch = 0 self.datas_in_batch = set() if done: xtl = (int(change.split('.')[-1]) - 1, 0) self.upd_stime(xtl) map(self.changelog_done_func, changes) self.archive_and_purge_changelogs(changes) break # it's either entry_ops() or Rsync that failed to do it's # job. Mostly it's entry_ops() [which currently has a problem # of failing to create an entry but failing to return an errno] # Therefore we do not know if it's either Rsync or the freaking # entry_ops() that failed... so we retry the _whole_ changelog # again. # TODO: remove entry retries when it's gets fixed. logging.warn('incomplete sync, retrying changelogs: %s' % ' '.join(map(os.path.basename, changes))) # Reset the Data counter before Retry self.status.dec_value("data", self.files_in_batch) self.files_in_batch = 0 time.sleep(0.5) def upd_stime(self, stime, path=None): if not path: path = self.FLAT_DIR_HIERARCHY if not stime == URXTIME: self.sendmark(path, stime) # Update last_synced_time in status file based on stime chkpt_time = gconf.configinterface.get_realtime( "checkpoint") checkpoint_time = 0 if chkpt_time is not None: checkpoint_time = int(chkpt_time) self.status.set_last_synced(stime, checkpoint_time) def update_worker_remote_node(self): node = sys.argv[-1] node_data = node.split("@") node = node_data[-1] remote_node_ip = node.split(":")[0] self.status.set_slave_node(remote_node_ip) def changelogs_batch_process(self, changes): changelogs_batches = [] current_size = 0 for c in changes: si = os.lstat(c).st_size if (si + current_size) > MAX_CHANGELOG_BATCH_SIZE: # Create new batch if single Changelog file greater than # Max Size! or current batch size exceeds Max size changelogs_batches.append([c]) current_size = si else: # Append to last batch, if No batches available Create one current_size += si if not changelogs_batches: changelogs_batches.append([c]) else: changelogs_batches[-1].append(c) for batch in changelogs_batches: logging.debug('processing changes %s' % repr(batch)) self.process(batch) def crawl(self): self.status.set_worker_crawl_status("Changelog Crawl") changes = [] # get stime (from the brick) and purge changelogs # that are _historical_ to that time. purge_time = self.get_purge_time() self.changelog_agent.scan() self.crawls += 1 changes = self.changelog_agent.getchanges() if changes: if purge_time: logging.info("slave's time: %s" % repr(purge_time)) processed = [x for x in changes if int(x.split('.')[-1]) < purge_time[0]] for pr in processed: logging.info( 'skipping already processed change: %s...' % os.path.basename(pr)) self.changelog_done_func(pr) changes.remove(pr) self.archive_and_purge_changelogs(processed) self.changelogs_batch_process(changes) def register(self, register_time, changelog_agent, status): self.changelog_agent = changelog_agent self.sleep_interval = int(gconf.change_interval) self.changelog_done_func = self.changelog_agent.done self.processed_changelogs_dir = os.path.join(self.setup_working_dir(), ".processed") self.name = "live_changelog" self.status = status class GMasterChangeloghistoryMixin(GMasterChangelogMixin): def register(self, register_time, changelog_agent, status): self.changelog_agent = changelog_agent self.changelog_register_time = register_time self.history_crawl_start_time = register_time self.changelog_done_func = self.changelog_agent.history_done self.history_turns = 0 self.processed_changelogs_dir = os.path.join(self.setup_working_dir(), ".history/.processed") self.name = "history_changelog" self.status = status def crawl(self): self.history_turns += 1 self.status.set_worker_crawl_status("History Crawl") purge_time = self.get_purge_time() end_time = int(time.time()) logging.info('starting history crawl... turns: %s, stime: %s, etime: %s' % (self.history_turns, repr(purge_time), repr(end_time))) if not purge_time or purge_time == URXTIME: logging.info("stime not available, abandoning history crawl") raise NoPurgeTimeAvailable() # Changelogs backend path is hardcoded as # /.glusterfs/changelogs, if user configured to different # location then consuming history will not work(Known issue as of now) changelog_path = os.path.join(gconf.local_path, ".glusterfs/changelogs") ret, actual_end = self.changelog_agent.history( changelog_path, purge_time[0], end_time, int(gconf.sync_jobs)) # scan followed by getchanges till scan returns zero. # history_scan() is blocking call, till it gets the number # of changelogs to process. Returns zero when no changelogs # to be processed. returns positive value as number of changelogs # to be processed, which will be fetched using # history_getchanges() while self.changelog_agent.history_scan() > 0: self.crawls += 1 changes = self.changelog_agent.history_getchanges() if changes: if purge_time: logging.info("slave's time: %s" % repr(purge_time)) processed = [x for x in changes if int(x.split('.')[-1]) < purge_time[0]] for pr in processed: logging.info('skipping already processed change: ' '%s...' % os.path.basename(pr)) self.changelog_done_func(pr) changes.remove(pr) self.changelogs_batch_process(changes) history_turn_time = int(time.time()) - self.history_crawl_start_time logging.info('finished history crawl syncing, endtime: %s, stime: %s' % (actual_end, repr(self.get_purge_time()))) # If TS returned from history_changelog is < register_time # then FS crawl may be required, since history is only available # till TS returned from history_changelog if actual_end < self.changelog_register_time: if self.history_turns < 2: sleep_time = 1 if history_turn_time < CHANGELOG_ROLLOVER_TIME: sleep_time = CHANGELOG_ROLLOVER_TIME - history_turn_time time.sleep(sleep_time) self.history_crawl_start_time = int(time.time()) self.crawl() else: # This exeption will be catched in resource.py and # fallback to xsync for the small gap. raise PartialHistoryAvailable(str(actual_end)) class GMasterXsyncMixin(GMasterChangelogMixin): """ This crawl needs to be xtime based (as of now it's not. this is because we generate CHANGELOG file during each crawl which is then processed by process_change()). For now it's used as a one-shot initial sync mechanism and only syncs directories, regular files, hardlinks and symlinks. """ XSYNC_MAX_ENTRIES = 1 << 13 def register(self, register_time=None, changelog_agent=None, status=None): self.status = status self.counter = 0 self.comlist = [] self.stimes = [] self.sleep_interval = 60 self.tempdir = self.setup_working_dir() self.tempdir = os.path.join(self.tempdir, 'xsync') self.processed_changelogs_dir = self.tempdir self.name = "xsync" logging.info('xsync temp directory: %s' % self.tempdir) try: os.makedirs(self.tempdir) except OSError: ex = sys.exc_info()[1] if ex.errno == EEXIST and os.path.isdir(self.tempdir): pass else: raise def crawl(self): """ event dispatcher thread this thread dispatches either changelog or synchronizes stime. additionally terminates itself on receiving a 'finale' event """ def Xsyncer(): self.Xcrawl() t = Thread(target=Xsyncer) t.start() logging.info('starting hybrid crawl..., stime: %s' % repr(self.get_purge_time())) self.status.set_worker_crawl_status("Hybrid Crawl") while True: try: item = self.comlist.pop(0) if item[0] == 'finale': logging.info('finished hybrid crawl syncing, stime: %s' % repr(self.get_purge_time())) break elif item[0] == 'xsync': logging.info('processing xsync changelog %s' % (item[1])) self.process([item[1]], 0) self.archive_and_purge_changelogs([item[1]]) elif item[0] == 'stime': logging.debug('setting slave time: %s' % repr(item[1])) self.upd_stime(item[1][1], item[1][0]) else: logging.warn('unknown tuple in comlist (%s)' % repr(item)) except IndexError: time.sleep(1) def write_entry_change(self, prefix, data=[]): if not getattr(self, "fh", None): self.open() self.fh.write("%s %s\n" % (prefix, ' '.join(data))) def open(self): try: self.xsync_change = os.path.join( self.tempdir, 'XSYNC-CHANGELOG.' + str(int(time.time()))) self.fh = open(self.xsync_change, 'w') except IOError: raise def close(self): if getattr(self, "fh", None): self.fh.flush() os.fsync(self.fh.fileno()) self.fh.close() self.fh = None def fname(self): return self.xsync_change def put(self, mark, item): self.comlist.append((mark, item)) def sync_xsync(self, last): """schedule a processing of changelog""" self.close() if self.counter > 0: self.put('xsync', self.fname()) self.counter = 0 if not last: time.sleep(1) # make sure changelogs are 1 second apart def sync_stime(self, stime=None, last=False): """schedule a stime synchronization""" if stime: self.put('stime', stime) if last: self.put('finale', None) def sync_done(self, stime=[], last=False): self.sync_xsync(last) if stime: # Send last as True only for last stime entry for st in stime[:-1]: self.sync_stime(st, False) if stime and stime[-1]: self.sync_stime(stime[-1], last) def is_sticky(self, path, mo): """check for DHTs linkto sticky bit file""" sticky = False if mo & 01000: sticky = self.master.server.linkto_check(path) return sticky def Xcrawl(self, path='.', xtr_root=None): """ generate a CHANGELOG file consumable by process_change. slave's xtime (stime) is _cached_ for comparisons across the filesystem tree, but set after directory synchronization. """ if path == '.': self.crawls += 1 if not xtr_root: # get the root stime and use it for all comparisons xtr_root = self.xtime('.', self.slave) if isinstance(xtr_root, int): if xtr_root != ENOENT: logging.warn("slave cluster not returning the " "correct xtime for root (%d)" % xtr_root) xtr_root = self.minus_infinity xtl = self.xtime(path) if isinstance(xtl, int): logging.warn("master cluster's xtime not found") xtr = self.xtime(path, self.slave) if isinstance(xtr, int): if xtr != ENOENT: logging.warn("slave cluster not returning the " "correct xtime for %s (%d)" % (path, xtr)) xtr = self.minus_infinity xtr = max(xtr, xtr_root) if not self.need_sync(path, xtl, xtr): if path == '.': self.sync_done([(path, xtl)], True) return self.xtime_reversion_hook(path, xtl, xtr) logging.debug("entering " + path) dem = self.master.server.entries(path) pargfid = self.master.server.gfid(path) if isinstance(pargfid, int): logging.warn('skipping directory %s' % (path)) for e in dem: bname = e e = os.path.join(path, e) xte = self.xtime(e) if isinstance(xte, int): logging.warn("irregular xtime for %s: %s" % (e, errno.errorcode[xte])) continue if not self.need_sync(e, xte, xtr): continue st = self.master.server.lstat(e) if isinstance(st, int): logging.warn('%s got purged in the interim ...' % e) continue if self.is_sticky(e, st.st_mode): logging.debug('ignoring sticky bit file %s' % e) continue gfid = self.master.server.gfid(e) if isinstance(gfid, int): logging.warn('skipping entry %s..' % e) continue mo = st.st_mode self.counter += 1 if ((stat.S_ISDIR(mo) or stat.S_ISLNK(mo) or stat.S_ISREG(mo))) else 0 if self.counter == self.XSYNC_MAX_ENTRIES: self.sync_done(self.stimes, False) self.stimes = [] if stat.S_ISDIR(mo): self.write_entry_change("E", [gfid, 'MKDIR', str(mo), str(0), str(0), escape(os.path.join(pargfid, bname))]) self.write_entry_change("M", [gfid, "SETATTR", str(st.st_uid), str(st.st_gid), str(st.st_mode), str(st.st_atime), str(st.st_mtime)]) self.Xcrawl(e, xtr_root) stime_to_update = xte # Live Changelog Start time indicates that from that time # onwards Live changelogs are available. If we update stime # greater than live_changelog_start time then Geo-rep will # skip those changelogs as already processed. But Xsync # actually failed to sync the deletes and Renames. Update # stime as min(Live_changelogs_time, Actual_stime) When it # switches to Changelog mode, it syncs Deletes and Renames. if self.live_changelog_start_time: stime_to_update = min(self.live_changelog_start_time, xte) self.stimes.append((e, stime_to_update)) elif stat.S_ISLNK(mo): self.write_entry_change( "E", [gfid, 'SYMLINK', escape(os.path.join(pargfid, bname))]) elif stat.S_ISREG(mo): nlink = st.st_nlink nlink -= 1 # fixup backend stat link count # if a file has a hardlink, create a Changelog entry as # 'LINK' so the slave side will decide if to create the # new entry, or to create link. if nlink == 1: self.write_entry_change("E", [gfid, 'MKNOD', str(mo), str(0), str(0), escape(os.path.join( pargfid, bname))]) else: self.write_entry_change( "E", [gfid, 'LINK', escape(os.path.join(pargfid, bname))]) self.write_entry_change("D", [gfid]) if path == '.': stime_to_update = xtl if self.live_changelog_start_time: stime_to_update = min(self.live_changelog_start_time, xtl) self.stimes.append((path, stime_to_update)) self.sync_done(self.stimes, True) class BoxClosedErr(Exception): pass class PostBox(list): """synchronized collection for storing things thought of as "requests" """ def __init__(self, *a): list.__init__(self, *a) # too bad Python stdlib does not have read/write locks... # it would suffivce to grab the lock in .append as reader, in .close as # writer self.lever = Condition() self.open = True self.done = False def wait(self): """wait on requests to be processed""" self.lever.acquire() if not self.done: self.lever.wait() self.lever.release() return self.result def wakeup(self, data): """wake up requestors with the result""" self.result = data self.lever.acquire() self.done = True self.lever.notifyAll() self.lever.release() def append(self, e): """post a request""" self.lever.acquire() if not self.open: raise BoxClosedErr list.append(self, e) self.lever.release() def close(self): """prohibit the posting of further requests""" self.lever.acquire() self.open = False self.lever.release() class Syncer(object): """a staged queue to relay rsync requests to rsync workers By "staged queue" its meant that when a consumer comes to the queue, it takes _all_ entries, leaving the queue empty. (I don't know if there is an official term for this pattern.) The queue uses a PostBox to accumulate incoming items. When a consumer (rsync worker) comes, a new PostBox is set up and the old one is passed on to the consumer. Instead of the simplistic scheme of having one big lock which synchronizes both the addition of new items and PostBox exchanges, use a separate lock to arbitrate consumers, and rely on PostBox's synchronization mechanisms take care about additions. There is a corner case racy situation, producers vs. consumers, which is not handled by this scheme: namely, when the PostBox exchange occurs in between being passed to the producer for posting and the post placement. But that's what Postbox.close is for: such a posting will find the PostBox closed, in which case the producer can re-try posting against the actual PostBox of the queue. To aid accumlation of items in the PostBoxen before grabbed by an rsync worker, the worker goes to sleep a bit after each completed syncjob. """ def __init__(self, slave, sync_engine, resilient_errnos=[]): """spawn worker threads""" self.log_err = False self.slave = slave self.lock = Lock() self.pb = PostBox() self.sync_engine = sync_engine self.errnos_ok = resilient_errnos for i in range(int(gconf.sync_jobs)): t = Thread(target=self.syncjob) t.start() def syncjob(self): """the life of a worker""" while True: pb = None while True: self.lock.acquire() if self.pb: pb, self.pb = self.pb, PostBox() self.lock.release() if pb: break time.sleep(0.5) pb.close() po = self.sync_engine(pb, self.log_err) if po.returncode == 0: ret = (True, 0) elif po.returncode in self.errnos_ok: ret = (False, po.returncode) else: po.errfail() pb.wakeup(ret) def add(self, e): while True: pb = self.pb try: pb.append(e) return pb except BoxClosedErr: pass def enable_errorlog(self): self.log_err = True def disable_errorlog(self): self.log_err = False