All new accounts created on Gitlab now require administrator approval. If you invite any collaborators, please let Flux staff know so they can approve the accounts.

loghole.in 70.2 KB
Newer Older
Timothy Stack's avatar
 
Timothy Stack committed
1 2 3 4
#! /usr/bin/env python

#
# EMULAB-COPYRIGHT
5
# Copyright (c) 2004, 2005, 2006 University of Utah and the Flux Group.
Timothy Stack's avatar
 
Timothy Stack committed
6 7 8 9 10 11
# All rights reserved.
#

import re
import pwd
import sys
Timothy Stack's avatar
 
Timothy Stack committed
12
import glob
Timothy Stack's avatar
 
Timothy Stack committed
13 14 15 16 17 18
import sets
import stat
import time
import getopt
import socket
import zipfile
19
import tempfile
Timothy Stack's avatar
 
Timothy Stack committed
20
import cStringIO
Timothy Stack's avatar
 
Timothy Stack committed
21
import threading
Timothy Stack's avatar
 
Timothy Stack committed
22 23 24 25 26 27 28 29 30 31 32
import xmlrpclib
import os, os.path
import ConfigParser

prefix = "@prefix@"

TBPATH = os.path.join(prefix, "lib")
if TBPATH not in sys.path:
    sys.path.append(TBPATH)
    pass

33 34
from tbevent import *

Timothy Stack's avatar
 
Timothy Stack committed
35 36 37 38 39 40 41
PACKAGE_VERSION = 0.1

# Default server
XMLRPC_SERVER = "@BOSSNODE@"
XMLRPC_PORT   = 3069

RSYNC = "@RSYNC@"
Timothy Stack's avatar
 
Timothy Stack committed
42
ACCEPTABLE_RSYNC_STATUS = [ 0 ]
43
NONRETRIABLE_RSYNC_STATUS = [ 0, 1, 2, 4 ]
Timothy Stack's avatar
 
Timothy Stack committed
44 45 46 47 48 49 50 51

DIRS = {
    "proj" : "proj",
    "groups" : "groups",
    }

EXPDIR_FMT = os.path.join("/", DIRS["proj"], "%(PID)s", "exp", "%(EID)s")

Timothy Stack's avatar
 
Timothy Stack committed
52 53
GLOBAL_LOGS = [
    "event-sched.log", "feedback.log", "../tbdata/feedback_data.tcl",
Timothy Stack's avatar
 
Timothy Stack committed
54
    "digest.out", "report.mail", "../tbdata/%(EID)s.report"
Timothy Stack's avatar
 
Timothy Stack committed
55
    ]
Timothy Stack's avatar
 
Timothy Stack committed
56

Timothy Stack's avatar
 
Timothy Stack committed
57 58
EXTENSIONS = [ ".zip" ]

Timothy Stack's avatar
 
Timothy Stack committed
59
LINK_DUMP_FMT = [ "trace_%s-%s.xmit", "trace_%s-%s.recv" ]
Timothy Stack's avatar
 
Timothy Stack committed
60

61 62
CLEANFILE = ".cleanonsync"

63 64 65
DOT_LOGHOLES = ""

DEFAULT_LOGHOLES = ["/var/emulab/logs", "/local/logs"]
Timothy Stack's avatar
 
Timothy Stack committed
66

67 68
DELAYLOG2XPLOT = os.path.join(prefix, "bin", "delaylog2xplot")

Timothy Stack's avatar
 
Timothy Stack committed
69 70 71
PID = None
EID = None

Timothy Stack's avatar
 
Timothy Stack committed
72 73 74 75
try:
    pw = pwd.getpwuid(os.getuid())
    pass
except KeyError:
Timothy Stack's avatar
 
Timothy Stack committed
76 77 78 79
    sys.stderr.write("error: unknown user id %d" % os.getuid())
    sys.exit(2)
    pass

Timothy Stack's avatar
 
Timothy Stack committed
80 81
USER = pw.pw_name
HOME = pw.pw_dir
Timothy Stack's avatar
 
Timothy Stack committed
82 83 84

CERTIFICATE = os.path.join(HOME, ".ssl", "emulab.pem")

85
PARALLELIZATION = 25
Timothy Stack's avatar
 
Timothy Stack committed
86

Timothy Stack's avatar
 
Timothy Stack committed
87 88 89 90 91 92 93 94 95
DEBUG = 0

VERBOSITY_MACHINE = 1
VERBOSITY_HUMAN = 2

VERBOSITY = VERBOSITY_HUMAN

SERVER = None

Timothy Stack's avatar
 
Timothy Stack committed
96 97
OSIDS = {}

98 99 100 101
JOB_TRIES = 3
JOB_PROGRESS = [ "-", "\\", "|", "/" ]
JOB_PROGRESS_INDEX = 0

Timothy Stack's avatar
 
Timothy Stack committed
102 103 104 105 106 107 108 109 110 111
if len(XMLRPC_SERVER) == 0:
    sys.stderr.write("internal error: XMLRPC_SERVER is empty\n")
    sys.exit(2)
    pass

if len(RSYNC) == 0:
    sys.stderr.write("internal error: RSYNC is empty\n")
    sys.exit(2)
    pass

112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135
IDENTITY = None

try:
    auth = open(os.path.join(HOME, ".ssh", "authorized_keys"), "r").read()

    for (pub, full) in ((os.path.join(HOME, ".ssh", "identity.pub"),
                         (1, os.path.join(HOME, ".ssh", "identity"))),
                        (os.path.join(HOME, ".ssh", "id_rsa.pub"),
                         (2, os.path.join(HOME, ".ssh", "id_rsa")))):
        try:
            idpub = open(pub, "r").read()
            
            if auth.find(idpub) != -1:
                IDENTITY = full
                break
            pass
        except IOError:
            pass
        pass
    pass
except IOError:
    pass


136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154
def loghole_include(path):
    retval = ""

    prefix = "/"
    for comp in path.split("/"):
        if comp != "":
            prefix = prefix + comp + "/"
            retval = retval + "+ " + prefix + "\n"
            pass
        pass

    retval = retval + "+ " + prefix + "**\n"
    
    return retval

for path in DEFAULT_LOGHOLES:
    DOT_LOGHOLES += loghole_include(path)
    pass

Timothy Stack's avatar
 
Timothy Stack committed
155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219
##
# Initialize the XML-RPC server proxy, if it has not already been done.  After
# returning, the SERVER variable will hold the proxy object.
#
def init_server():
    global SERVER

    if not SERVER:
        if not os.path.exists(CERTIFICATE):
            sys.stderr.write("error: missing emulab certificate: %s\n" %
                             CERTIFICATE)
            sys.exit(2)
            pass
        
        from M2Crypto.m2xmlrpclib import SSL_Transport
        from M2Crypto import SSL
        
        URI = "https://" + XMLRPC_SERVER + ":" + str(XMLRPC_PORT) + prefix
        
        ctx = SSL.Context("sslv23")
        ctx.load_cert(CERTIFICATE, CERTIFICATE)
        ctx.set_verify(SSL.verify_none, 16)
        ctx.set_allow_unknown_ca(0)
        
        SERVER = xmlrpclib.ServerProxy(URI, SSL_Transport(ctx))
        pass
    return

##
# Download the user's experiment list from the XML-RPC server.
#
# @return A list of (pid, eid) pairs for each experiment created by the user.
#
def get_explist():
    global SERVER

    from emulabclient import RESPONSE_SUCCESS
    
    retval = []
    
    init_server()
    
    list_method = getattr(SERVER, "experiment.getlist")
    elist = list_method(PACKAGE_VERSION, {})
    
    if elist["code"] != RESPONSE_SUCCESS:
        sys.stderr.write("error: " + elist["output"] + "\n")
        sys.exit(1)
        pass
    else:
        for project in elist["value"]:
            for group in elist["value"][project]:
                for exp in elist["value"][project][group]:
                    retval.append((project, exp))
                    pass
                pass
            pass
        pass
    
    return retval

##
# Print the top level usage statement for the whole tool.
#
def usage():
Timothy Stack's avatar
 
Timothy Stack committed
220
    print "Usage: loghole [-hV] [-e pid/eid|-a] [OPTIONS] <action> [...]"
Timothy Stack's avatar
 
Timothy Stack committed
221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242
    print
    print "Manage an experiment's log hole."
    print
    print "Optional arguments:"
    print "  -h, --help                Print this message, or the action's"
    print "                            usage message."
    print "  -V, --version             Print version information."
    print "  -d, --debug               Turn on debugging messages."
    print "  -q, --quiet               Be less verbose."
    print "  -v, --verbose             Be more verbose."
    print "  -e, --experiment=PID/EID  Specify the experiment to manage."
    print "  -a, --all                 Perform the action on all of your "
    print "                            experiments."
    print "  -s, --server=SERVER       Specify the XML-RPC server to use."
    print "                            (Default: %s)" % XMLRPC_SERVER
    print "  -P, --port=PORT           Specify the XML-RPC server port."
    print "                            (Default: %d)" % XMLRPC_PORT
    print
    print "Actions:"
    print "  sync      Synchronize the experiment's log holes with the"
    print "            experiment's log directory."
    print "  archive   Archive the contents of the experiment's log directory."
Timothy Stack's avatar
 
Timothy Stack committed
243
    print "  change    Change the metadata of an existing archive."
Timothy Stack's avatar
 
Timothy Stack committed
244 245 246 247
    print "  list      Print a brief listing of the archives in the"
    print "            experiment's log directory."
    print "  show      Show a detailed listing of the archive(s) in the"
    print "            experiment directory."
248
    print "  validate  Validate the experiment logs."
Timothy Stack's avatar
 
Timothy Stack committed
249 250
    print "  clean     Clean the experiment log directory and node local"
    print "            directories."
Timothy Stack's avatar
 
Timothy Stack committed
251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266
    print "  gc        Garbage collect old archives."
    print
    print "Environment:"
    print "  cwd       The project and experiment ID will be inferred from the"
    print "            current working directory, if it is inside the"
    print "            experiment's directory (e.g. /proj/foo/exp/bar)."
    print
    print "Examples:"
    print "  $ loghole -e foo/bar sync"
    print "  $ loghole -e foo/bar list"
    return

##
# Print the usage statement for the "sync" action.
#
def sync_usage():
267
    print "Usage: loghole sync [-nPs] [-r remotedir] [-l localdir] [node1 ...]"
Timothy Stack's avatar
 
Timothy Stack committed
268
    print
269
    print "Synchronize the experiment's log holes with a local directory."
Timothy Stack's avatar
 
Timothy Stack committed
270 271
    print
    print "Optional arguments:"
272 273 274 275 276 277 278 279 280 281 282
    print "  -r, --remote=REMOTEDIR"
    print "               An additional remote directory to sync.  This option"
    print "               is additive, so you can download several additional"
    print "               directories."
    print "  -l, --local=LOCALDIR"
    print "               The local directory to store the downloaded files."
    print "               This defaults to the experiment's log directory."
    print "  -n, --no-standard"
    print "               Flag that indicates that the standard logholes"
    print "               (i.e. '/var/emulab/logs', '/local/logs') should not"
    print "               be downloaded."
283 284
    print "  -P, --no-post"
    print "               Do not do any additional post processing."
285 286
    print "  -s, --snapshot"
    print "               Send a SNAPSHOT event to any link trace agents."
Timothy Stack's avatar
 
Timothy Stack committed
287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323
    print "  node1 ...    The names of the virtual or physical nodes that "
    print "               should be synced.  The default behavior is to sync "
    print "               all of the nodes in an experiment."
    print
    print "Files:"
    print "  /proj/<pid>/exp/<eid>/logs/<node>/<loghole>"
    print "               The location of the log hole for a particular node."
    print
    print "Examples:"
    print "  $ loghole sync"
    return

##
# Print the usage statement for the "archive" action.
#
def archive_usage():
    print ("Usage: loghole archive [-k (i-delete|space-is-needed)] "
           "[-a days] [-c comment] [-d] [archive-name]")
    print
    print "Archive the contents of the experiment's log directory.  The"
    print "archive will be a zip file containing all of the directories"
    print "under the experiment's log directory, any extra log files"
    print "at the top (e.g. event-sched.log), and some metadata."
    print
    print "Optional arguments:"
    print "  -k, --keep-until=(i-delete|space-is-needed)"
    print "                         Keep the archive until you decide to"
    print "                         delete it manually or space is needed."
    print "                         (Default: space-is-needed)"
    print "  -a, --keep-atleast=N   Keep the archive atleast N days after"
    print "                         creation.  This value keeps the archive "
    print "                         from being garbage collected when more "
    print "                         space is needed. (Default: 3 days)"
    print "  -c, --comment=COMMENT  Add a comment to the archive. This comment"
    print "                         will be displayed in the listing generated"
    print "                         by the 'show' action.  Use a single dash "
    print "                         (-) to read the comment from standard in."
324 325
    print "  -d, --delete           Mark the experiment as clean-on-sync to"
    print "                         remove stale data before another sync."
Timothy Stack's avatar
 
Timothy Stack committed
326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395
    print
    print "Required arguments:"
    print "  archive-name  The name of the archive to create."
    print "                (Default: The experiment name)"
    print
    print "Files:"
    print "  /proj/<pid>/exp/<eid>/logs/<archive-name>.<instance>.zip"
    print
    print "Examples:"
    print "  $ loghole archive foobar"
    return

##
# Print the usage statement for the "list" action.
#
def list_usage():
    print "Usage: loghole list [-O1!Xo] [-m atmost] [-s megabytes]"
    print
    print "Print a brief listing of the archives in the experiment's log"
    print "directory.  The listing can be helpful in getting an idea of"
    print "what archives are in your experiment and when they are likely"
    print "to be garbage collected."
    print
    print "Optional arguments:"
    print "  -O                    Only list archives that are marked as 'keep"
    print "                        until \"i-delete\"'."
    print "  -1                    Only list archives that are a day from"
    print "                        their keep-atleast date."
    print "  -!                    Only list archives that are past their"
    print "                        keep-atleast date."
    print "  -X                    Only list archives that are ready to be"
    print "                        garbage collected."
    print "  -o                    List archives that do not match the above"
    print "                        flags."
    print "  -m, --keep-atmost=N   Keep at most N archives in the experiment."
    print "                        (Default: 100)"
    print "  -s, --keep-size=N     Keep at N megabytes of archives."
    print "                        (Default: 3.0 MB)"
    print
    print "Examples:"
    print "  $ loghole list"
    print "  GC  Name             Date"
    print
    print "  [ ] foobar.1.zip     10/15"
    print "  [!] foobar.0.zip     10/13"
    print "  ..."
    print
    print "  $ loghole -q list -!"
    print "  foobar.0.zip"
    return

##
# Print the usage statement for the "show" action.
#
def show_usage():
    print "Usage: loghole show [archive-name]"
    print
    print "Show a detailed listing of the archive(s) in the experiment"
    print "directory.  The listing shows all of the metadata for the archive,"
    print "comments, and the file listing."
    print
    print "Example:"
    print "  $ loghole show foobar"
    print "  Archive: foobar.0.zip"
    print "    Creator:\t\t%s" % USER
    print "    ..."
    print "  Archive: foobar.1.zip"
    print "    ..."
    return

Timothy Stack's avatar
 
Timothy Stack committed
396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432
##
# Print the usage statement for the "change" action.
#
def change_usage():
    print ("Usage: loghole change [-k (i-delete|space-is-needed)] "
           "[-a days] [-c comment] <archive-name1> [archive-name2 ...]")
    print
    print "Change the metadata of an existing archive."
    print
    print "Optional arguments:"
    print "  -k, --keep-until=(i-delete|space-is-needed)"
    print "                         Keep the archive until you decide to"
    print "                         delete it manually or space is needed."
    print "                         (Default: space-is-needed)"
    print "  -a, --keep-atleast=N   Keep the archive atleast N days after"
    print "                         creation.  This value keeps the archive "
    print "                         from being garbage collected when more "
    print "                         space is needed. (Default: 3 days)"
    print "  -c, --comment=COMMENT  Add a comment to the archive. This comment"
    print "                         will be displayed in the listing generated"
    print "                         by the 'show' action.  Use a single dash "
    print "                         (-) to read the comment from standard in."
    print
    print "Required arguments:"
    print "  archive-name  The name of the archive to change."
    print
    print "Files:"
    print "  /proj/<pid>/exp/<eid>/logs/<archive-name>.<instance>.zip"
    print
    print "Examples:"
    print "  $ loghole change -c 'Results bad...' foobar"
    return

##
# Print the usage statement for the "validate" action.
#
def validate_usage():
433
    print "Usage: loghole validate"
Timothy Stack's avatar
 
Timothy Stack committed
434
    print
435
    print "Validate the experiment logs.  Currently, this action will:"
Timothy Stack's avatar
 
Timothy Stack committed
436
    print
437 438 439
    print "  1. Examine the log files from individual logs from the "
    print "     program-agent to verify that the files are intact."
    print "  2. Check that the referent of any soft-links exist."
Timothy Stack's avatar
 
Timothy Stack committed
440 441
    print
    print "Examples:"
442
    print "  $ loghole validate"
Timothy Stack's avatar
 
Timothy Stack committed
443 444
    return

Timothy Stack's avatar
 
Timothy Stack committed
445 446 447 448
##
# Print the usage statement for the "clean" action.
#
def clean_usage():
449
    print "Usage: loghole clean [-fner] [node1 ...]"
Timothy Stack's avatar
 
Timothy Stack committed
450 451 452 453
    print
    print "Clean the experiment log directory by removing any subdirectories"
    print "and/or clean the log holes on the nodes.  The default behavior is"
    print "to clean both."
Timothy Stack's avatar
 
Timothy Stack committed
454
    print
Timothy Stack's avatar
 
Timothy Stack committed
455 456 457 458
    print "Optional arguments:"
    print "  -f, --force         Do not ask for confirmation."
    print "  -n, --nodes         Clean only the nodes log directories."
    print "  -e, --experiment    Clean only the experiment log directory."
459
    print "  -r, --root          Use sudo to clean nodes as root."
Timothy Stack's avatar
 
Timothy Stack committed
460 461 462 463 464
    print
    print "Examples:"
    print "  $ loghole clean"
    return

Timothy Stack's avatar
 
Timothy Stack committed
465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480
##
# Print the usage statement for the "gc" action.
#
def gc_usage():
    print "Usage: loghole gc [-n] [-m atmost] [-s megabytes]"
    print
    print "Garbage collect old log archives."
    print
    print "Optional arguments:"
    print "  -n, --dry-run         Perform a dry run, do not delete any files."
    print "  -m, --keep-atmost=N   Keep at most N archives in the experiment."
    print "                        (Default: 100)"
    print "  -s, --keep-size=N     Keep at N megabytes of archives."
    print "                        (Default: 3.0 MB)"
    print
    print "Examples:"
Timothy Stack's avatar
 
Timothy Stack committed
481
    print "  $ loghole gc -n -m 1"
Timothy Stack's avatar
 
Timothy Stack committed
482 483 484
    print "  Reclaimed 3421 bytes from foo/bar."
    print "  52434 bytes are still in foo/bar."
    print
Timothy Stack's avatar
 
Timothy Stack committed
485
    print "  $ loghole -q gc -m 1"
Timothy Stack's avatar
 
Timothy Stack committed
486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529
    print "  3421 52434"
    return

##
# Infer the project and experiment IDs from the given list of directories.
#
# @param *args The list of directories to check for likely PID/EID values.
# @return A pair containing the project and experiment IDs, which will be None
# if the IDs could not be inferred.
#
def infer_pid_eid(*args):
    pid = None
    eid = None
    for path in args:
        dirs = path.split(os.path.sep)
        if ((len(dirs) < 6) or not
            ((DIRS["proj"] in dirs and
              "exp" in dirs[dirs.index(DIRS["proj"]):]) or
             (len(dirs) >= 6 and
              (DIRS["groups"] in dirs and
               "exp" in dirs[dirs.index(DIRS["groups"]):])))):
            # No match...
            pass
        else:
            if DIRS["proj"] in dirs:
                pid = dirs[dirs.index(DIRS["proj"]) + 1]
                pass
            else:
                pid = dirs[dirs.index(DIRS["groups"]) + 1]
                pass
            eid = dirs[dirs.index("exp") + 1]
            pass
        pass

    return pid, eid

##
# Attempt to open a loghole archive located at the given path.  An archive
# must be a zip file with a "loghole.cfg" file stored in the root directory.
#
# @param path The path to the loghole archive.
# @return A pair containing the ZipFile object referring to the archive and a
# ConfigParser object that contains the archive's metadata.
#
Timothy Stack's avatar
 
Timothy Stack committed
530
def open_loghole(path, mode='r'):
Timothy Stack's avatar
 
Timothy Stack committed
531 532
    retval = None, None
    if zipfile.is_zipfile(path):
Timothy Stack's avatar
 
Timothy Stack committed
533
        zf = zipfile.ZipFile(path, mode)
Timothy Stack's avatar
 
Timothy Stack committed
534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782
        if "loghole.cfg" in zf.namelist():
            cio = cStringIO.StringIO(zf.read("loghole.cfg"))
            cp = ConfigParser.ConfigParser()
            cp.readfp(cio)
            retval = zf, cp
            zf = None
            pass
        if zf:
            zf.close()
            pass
        pass
    
    return retval

##
# Generate a name that is unique in the context of the given list of entries.
#
# @param base The base of the file name.
# @param ext The file name extension.
# @param entries An iterable containing any existing file names.
# @return A string of the form "<base>.<unique #>.<ext>".
#
def uniquish_name(base, ext, entries):
    retval = None
    lpc = 0
    while not retval:
        name = base + "." + str(lpc) + "." + ext
        if name not in entries:
            retval = name
            pass
        lpc = lpc + 1
        pass
    
    return retval

##
# Callback for os.walk() that writes files to the ZipFile.
#
# @param zf The ZipFile object that files should be added to.
# @param dirname The current directory name.
# @param names The list of files in the current directory.
#
def file_archiver(zf, dirname, names):
    for entry in names:
        entry = os.path.join(dirname, entry)
        if os.path.isfile(entry):
            zf.write(entry)
            pass
        pass

    return

##
# Sort the list of log archives in the current directory based on when they
# should be garbage collected.
#
# @param keep_atmost The maximum number of archives that should be kept, any
# extras are marked as collectable.
# @param keep_size The number of megabytes of archives to keep.
# @return A dictionary containing: "del", a list of archives that will be
# deleted at the next GC; "safe", a list of archives that are marked "keep
# until 'i-delete'"; "soon", the remaining archives; "reclaimed-size", the
# amount of data that will be reclaimed at the next GC; and "inuse-size", the
# amount of data that will remain after the next GC.
#
def sort_archives(keep_atmost, keep_size):
    
    keep_size = keep_size * 1024 * 1024

    reclaimed_size = 0
    inuse_size = 0 # In-use size for non-gc'able archives
    inuse_gc_size = 0 # In-use size for gc'able archives
    
    current_time = time.time()
    
    retval_del = []
    retval_safe = []
    retval_soon = []
    exp_logholes = []

    # Walk the directory collecting archives and sorting them into "safe" and
    # to-be-determined lists.
    for entry in os.listdir(os.curdir):
        st = os.stat(entry)
        lh, cp = open_loghole(entry)
        if lh:
            keep_until = cp.get("MAIN", "keep-until")
            if keep_until == "i-delete":
                if DEBUG:
                    print "loghole: ignoring - " + entry
                    pass

                retval_safe.append(entry)
                inuse_size = inuse_size + st[stat.ST_SIZE]
                pass
            else:
                created = time.mktime(time.strptime(cp.get("MAIN", "created")))
                keep_atleast = (cp.getint("MAIN", "keep-atleast") *
                                60 * 60 * 24)

                inuse_gc_size = inuse_gc_size + st[stat.ST_SIZE]
                exp_logholes.append((entry, st, created + keep_atleast))
                pass
            lh.close()
            pass
        pass

    # Sort the archives from oldest keep-atleast time to newest, then
    exp_logholes.sort(lambda x, y: cmp(x[2], y[2]))

    # ... divide them between the del list and the soon list depending on
    # whether or not space is needed.
    for entry, st, expiration in exp_logholes:
        if inuse_gc_size > keep_size:
            if DEBUG:
                print "loghole: stale entry - " + entry
                pass

            retval_del.append(entry)
            inuse_gc_size = inuse_gc_size - st[stat.ST_SIZE]
            reclaimed_size = reclaimed_size + st[stat.ST_SIZE]
            pass
        else:
            retval_soon.append(entry)
            pass
        pass

    # Resort the list in the reverse direction for keep-atleast time.
    exp_logholes = [x for x in exp_logholes if x[0] in retval_soon]
    exp_logholes.sort(lambda x, y: cmp(y[2], x[2]))

    # Update the inuse_size for the archives below the keep-atmost value, then
    for entry, st, expiration in exp_logholes[0:keep_atmost]:
        if DEBUG:
            print ("loghole: keeping '"
                   + entry
                   + "' until "
                   + time.ctime(expiration))
            pass

        inuse_size = inuse_size + st[stat.ST_SIZE]
        pass

    # ... move the rest to the "del" list and
    for entry, st, expiration in exp_logholes[keep_atmost:]:
        if DEBUG:
            print "loghole: excess entry - " + entry
            pass

        retval_soon.remove(entry)
        retval_del.append(entry)
        reclaimed_size = reclaimed_size + st[stat.ST_SIZE]
        pass

    # ... drop the end of the array.
    del retval_soon[keep_atmost:]

    return {
        "del" : retval_del,
        "safe" : retval_safe,
        "soon" : retval_soon,
        "reclaimed-size" : reclaimed_size,
        "inuse-size" : inuse_size,
        }

##
# Performs the "archive" action, which involves building a zip file containing
# all of the logs.
#
# @param args Action-specific command line arguments.
#
def do_archive(args):
    retval = 0

    keep_until = "space-is-needed"
    keep_atleast = 3
    comments = []
    delete_dir = False

    # Parse action-specific options and arguments.
    try:
        opts, args = getopt.getopt(args, "k:a:c:d", [
            "keep-until=", "keep-atleast=", "comment=", "delete" ])
        for opt, val in opts:
            if opt in ("-k", "--keep-until"):
                if val == "i-delete" or val == "space-is-needed":
                    keep_until = val
                    pass
                else:
                    raise getopt.error(
                        'error: expecting "i-delete" or "space-is-needed" for '
                        'keep until value, got - ' + val)
                pass
            elif opt in ("-a", "--keep-atleast"):
                try:
                    keep_atleast = int(val)
                    pass
                except:
                    raise getopt.error(
                        'error: expecting integer for keep atleast value, '
                        'got - ' + val)
                pass
            elif opt in ("-c", "--comment"):
                if val == "-":
                    comments.append(sys.stdin.read())
                    pass
                else:
                    comments.append(val)
                    pass
                pass
            elif opt in ("-d", "--delete"):
                delete_dir = True
                pass
            pass
        
        if len(args) > 1:
            raise getopt.error(
                'error: too many arguments for "archive" action.')
        pass
    except getopt.error, e:
        print e.args[0]
        archive_usage()
        return 2

    if len(args) < 1:
        archive_base = EID
        pass
    else:
        archive_base = args[0]
        pass

    logdir = os.path.join(EXPDIR_FMT % { "PID" : PID, "EID" : EID }, "logs")
    os.chdir(logdir)

    # Figure out a unique name for the archive,
    archive_name = uniquish_name(archive_base, 'zip', os.listdir(logdir))

    # ... create a new zip file,
    zf = zipfile.ZipFile(archive_name, 'w', zipfile.ZIP_DEFLATED)

    # ... fill it up with the node log holes, and
    for entry in os.listdir(logdir):
        if os.path.isdir(entry):
            os.path.walk(entry, file_archiver, zf)
            pass
        pass

    # ... any global log files (e.g. event-sched.log).
    for log in GLOBAL_LOGS:
Timothy Stack's avatar
 
Timothy Stack committed
783 784 785
        formatted_log = log % { "PID" : PID, "EID" : EID }
        if os.path.exists(formatted_log):
            zf.write(formatted_log)
Timothy Stack's avatar
 
Timothy Stack committed
786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832
            pass
        pass

    # Construct the meta-data.
    cp = ConfigParser.ConfigParser()
    cp.add_section("MAIN")
    cp.set("MAIN", "created", time.ctime(time.time()))
    cp.set("MAIN", "creator", USER)
    cp.set("MAIN", "project", PID)
    cp.set("MAIN", "experiment", EID)
    cp.set("MAIN", "keep-until", keep_until)
    cp.set("MAIN", "keep-atleast", keep_atleast)
    
    cio = cStringIO.StringIO()
    cp.write(cio)

    loghole_zi = zipfile.ZipInfo()
    loghole_zi.filename = "loghole.cfg"
    ct = time.localtime(time.time())
    loghole_zi.date_time = (ct[0], ct[1], ct[2], ct[3], ct[4], ct[5])
    loghole_zi.compress_type = zipfile.ZIP_STORED
    loghole_zi.comment = "Log hole meta-data"
    
    zf.writestr(loghole_zi, cio.getvalue())

    # Add any comments.
    ct = time.localtime(time.time())
    lpc = 0
    for comment in comments:
        loghole_zi = zipfile.ZipInfo()
        loghole_zi.filename = "loghole-comment.%d.txt" % lpc
        loghole_zi.date_time = (ct[0], ct[1], ct[2], ct[3], ct[4], ct[5])
        loghole_zi.compress_type = zipfile.ZIP_DEFLATED
        loghole_zi.comment = "Loghole comment #%d" % lpc

        zf.writestr(loghole_zi, comment)

        lpc = lpc + 1
        pass
    
    if DEBUG:
        zf.printdir()
        pass
    
    zf.close()

    if delete_dir:
833
        open(CLEANFILE, 'w')
834
        os.chmod(CLEANFILE, 0664)
Timothy Stack's avatar
 
Timothy Stack committed
835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869
        pass

    if VERBOSITY == VERBOSITY_MACHINE:
        print archive_name
        pass
    elif VERBOSITY >= VERBOSITY_HUMAN:
        print "Created log hole archive: " + archive_name
        pass
    
    return retval

##
# Performs the "show" action, which shows a detailed listing of a log archive.
#
# @param args Action-specific command line arguments.
#
def do_show(args):
    retval = 0

    logdir = os.path.join(EXPDIR_FMT % { "PID" : PID, "EID" : EID }, "logs")
    os.chdir(logdir)
    
    for entry in os.listdir(logdir):
        if len(args) == 0 or entry.startswith(args[0]):
            lh, cp = open_loghole(entry)
            if lh:
                print "Archive: " + entry
                print ("  Project/Experiment:\t%s/%s" %
                       (cp.get("MAIN", "project"),
                        cp.get("MAIN", "experiment")))
                print "  Creator:\t\t%s" % cp.get("MAIN", "creator")
                print "  Created:\t\t%s" % cp.get("MAIN", "created")
                print "  Keep-until:\t\t%s" % cp.get("MAIN", "keep-until")
                print ("  Keep-atleast:\t\t%s days" %
                       cp.get("MAIN", "keep-atleast"))
Timothy Stack's avatar
 
Timothy Stack committed
870 871 872 873
                if "report.mail" in lh.namelist():
                    print "Report:"
                    print lh.read("report.mail")
                    pass
Timothy Stack's avatar
 
Timothy Stack committed
874 875 876 877 878 879 880 881 882
                for name in lh.namelist():
                    if re.match(r'loghole-comment.*\.txt', name):
                        before = 38 - (len(name) / 2)
                        after = 77 - before - len(name)
                        print "%s %s %s" % ("-" * before, name, "-" * after)
                        print lh.read(name)
                        pass
                    pass
                print "=" * 79
Timothy Stack's avatar
 
Timothy Stack committed
883 884 885 886 887
                try:
                    lh.printdir()
                    pass
                except IOError:
                    pass
Timothy Stack's avatar
 
Timothy Stack committed
888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903
                lh.close()
                print
                pass
            pass
        pass
    
    return retval

##
# Run rsync for a given host and list of files.
#
# @param host The host to sync against.
# @param base The local directory where the downloaded files should be placed.
# @param *args The list of directories to sync.
# @retval The exit status of the rsync command-line.
#
904
def rsync(host, base, dlpath, paths = ["/"], port = None, output = False,
905
          delay = 0, as_root = False):
Timothy Stack's avatar
 
Timothy Stack committed
906 907 908
    global RSYNC, HOME, DEBUG
    
    retval = 0
909
    for path in paths:
Timothy Stack's avatar
 
Timothy Stack committed
910 911 912 913 914 915
        fullpath = os.path.join("", base, path[1:], "")
        try:
            os.makedirs(fullpath)
            pass
        except OSError:
            pass
Timothy Stack's avatar
 
Timothy Stack committed
916
        cmd = [RSYNC,
917
               "-rtgoDlz",
918
               "-v", # XXX
Timothy Stack's avatar
 
Timothy Stack committed
919
               "--copy-unsafe-links",
920
               "--include-from=" + dlpath]
Timothy Stack's avatar
 
Timothy Stack committed
921
        if VERBOSITY > VERBOSITY_HUMAN:
Timothy Stack's avatar
 
Timothy Stack committed
922
            cmd.append("-v")
Timothy Stack's avatar
 
Timothy Stack committed
923
            pass
Timothy Stack's avatar
Timothy Stack committed
924
        ssh_args = ("ssh -o BatchMode=yes -o StrictHostKeyChecking=no "
925 926
                    + "-o UserKnownHostsFile="
                    + os.path.join(HOME, ".ssh", "emulab_known_hosts"))
927 928 929
        if IDENTITY:
            ssh_args += " -o Protocol=%d -i %s" % IDENTITY
            pass
930 931 932 933
        if port:
            ssh_args += " -p " + str(port)
            pass
        cmd.extend([("--rsh=" + ssh_args),
Timothy Stack's avatar
 
Timothy Stack committed
934 935
                    host + ":" + path,
                    fullpath])
Timothy Stack's avatar
 
Timothy Stack committed
936
        if DEBUG:
Timothy Stack's avatar
 
Timothy Stack committed
937 938
            print `cmd`
            retval = -1
Timothy Stack's avatar
 
Timothy Stack committed
939 940
            pass
        else:
Timothy Stack's avatar
 
Timothy Stack committed
941 942 943 944 945
            retval = os.fork()
            if retval != 0:
                # Parent
                pass
            else:
946
                try:
947 948 949
                    if delay > 0:
                        time.sleep(delay)
                        pass
950 951 952 953 954 955 956 957 958
                    os.close(0)
                    os.open("/dev/null", os.O_RDONLY)
                    if output:
                        os.close(1)
                        fd = os.open(os.path.join(base, ".rsync.out"),
                                     os.O_CREAT|os.O_TRUNC|os.O_WRONLY,
                                     0664)
                        os.dup2(fd, 2)
                        pass
959 960 961 962 963 964 965 966
                    try:
                        os.execve(RSYNC, cmd, os.environ)
                    except OSError, e:
                        sys.stderr.write(
                            "error: unable to exec '%s' - %s\n"
                            "error: args=%s\n"
                            % (RSYNC, e.strerror, `cmd`))
                        pass
967 968 969 970 971
                    os._exit(127)
                    pass
                except:
                    sys.exit(1)
                    pass
Timothy Stack's avatar
 
Timothy Stack committed
972
                pass
Timothy Stack's avatar
 
Timothy Stack committed
973 974 975 976 977
            pass
        pass
    
    return retval

978
def rmstar(host, port, delay, as_root, *args):
Timothy Stack's avatar
 
Timothy Stack committed
979 980 981 982 983
    global HOME, DEBUG

    retval = 0
    for path in args:
        cmd = ["/usr/bin/ssh",
Timothy Stack's avatar
Timothy Stack committed
984
               "-o", "BatchMode=yes",
Timothy Stack's avatar
 
Timothy Stack committed
985 986
               "-o", "StrictHostKeyChecking=no",
               "-o", ("UserKnownHostsFile=" +
Timothy Stack's avatar
Timothy Stack committed
987 988 989 990
                      os.path.join(HOME, ".ssh", "emulab_known_hosts"))]
        if port:
            cmd.extend([ "-p", str(port) ])
            pass
991 992 993 994
        cmd.extend([ host ])
        if as_root:
            cmd.extend(["sudo"])
        cmd.extend(["rm", "-rf", path ])
Timothy Stack's avatar
 
Timothy Stack committed
995 996 997 998 999 1000 1001 1002 1003 1004
        if DEBUG:
            print `cmd`
            retval = -1
            pass
        else:
            retval = os.fork()
            if retval != 0:
                # Parent
                pass
            else:
1005 1006 1007
                if delay > 0:
                    time.sleep(delay)
                    pass
Timothy Stack's avatar
 
Timothy Stack committed
1008 1009 1010 1011 1012 1013 1014
                os.execve("/usr/bin/ssh", cmd, os.environ)
                os._exit(127)
                pass
            pass
        pass
    
    return retval
1015

1016 1017 1018 1019 1020 1021 1022 1023 1024
def print_status(msg):
    if not sys.stdout.isatty() or VERBOSITY < VERBOSITY_HUMAN:
        return
    
    sys.stdout.write(msg)
    sys.stdout.flush()
    
    return

1025
def print_job_status(mode, hosts, jobs, hoststatus, hoststart):
1026 1027
    if not sys.stdout.isatty() or VERBOSITY < VERBOSITY_HUMAN:
        return
Timothy Stack's avatar
 
Timothy Stack committed
1028
    
1029
    lmsg = "nodes todo/doing/done %d/%d/%d "
1030
    rmsg = mode + " time min/avg/max: %.2f/%.2f/%.2f secs"
1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050

    hosttimes = map(lambda x:
                    hoststatus.get(x, (time.time(),))[0] - hoststart[x],
                    hoststart.keys())
    
    if len(hosttimes) == 0:
        hosttimes = [0]
        avg = 0.0
        pass
    else:
        avg = sum(hosttimes) / len(hosttimes)
        pass
    lmsg = lmsg % (
        len(hosts) - len(hoststatus) - len(jobs),
        len(jobs),
        len(hoststatus))
    rmsg = rmsg % (
        min(hosttimes),
        avg,
        max(hosttimes))
1051
    line_len = len(lmsg) + len(rmsg) + 2
1052 1053 1054
    if line_len < 79:
        lmsg += " " * (79 - line_len)
        pass
1055 1056
    print_status(lmsg + rmsg + " "
                 + JOB_PROGRESS[JOB_PROGRESS_INDEX % len(JOB_PROGRESS)] + "\r")
1057 1058

    return
Timothy Stack's avatar
 
Timothy Stack committed
1059

1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084
##
# Detect changes in the rsync log file sizes.
#
# @param jobs The hashtable of jobs
# @param logsize The hashtable of previous log file sizes.
# @param logdir The root directory for log files.
# @return True if the size of any of the log files has changed.
#
def making_progress(jobs, logsize, logdir):
    retval = False
    for pid in jobs:
        phost, vhost = jobs[pid]
        try:
            st = os.stat(os.path.join(logdir, vhost, ".rsync.out"))
            if st[stat.ST_SIZE] > logsize[pid]:
                retval = True
                logsize[pid] = st[stat.ST_SIZE]
                pass
            pass
        except OSError:
            pass
        pass
    
    return retval

1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129
##
# Trace preprocessing.
#
# @param traces Sequence of dictionaries describing the active link tracers.
#
def trace_pre(traces):
    at = address_tuple()
    at.eventtype = "SNAPSHOT"
    at.expt = "%s/%s" % (PID, EID)
    at.scheduler = 1

    tv = timeval()
    tv.tv_sec = int(time.time())
    
    notified = {}
    ec = EventClient(server="event-server",
                     keyfile=os.path.join("/proj", PID, "exp", EID,
                                          "tbdata", "eventkey"))

    for trace in traces:
        if trace["trace_type"] == "monitor":
            continue
        
        lvn = trace["linkvname"]
        if lvn not in notified:
            at.objname = "%s-tracemon" % (lvn,)
            en = ec.create_notification(at)
            ec.schedule(en, tv)
            notified[lvn] = 1
            pass
        
        pass

    if len(notified) > 0:
        time.sleep(1)
        pass
    
    return

##
# Trace post processing
#
# @param link2trace Dictionary containing the link names that map to a sequence
# of dictionaries containing individual trace information.
#
1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151
def trace_post(link2trace):
    lpc = 1
    for linkvname in link2trace.keys():
        print_status("linktrace post processing %d/%d  " %
                     (lpc, len(link2trace)))
        args = {}
        os.chdir(linkvname)
        for filename in glob.glob("*.recv*"):
            base, ext = os.path.splitext(filename)
            if base.startswith("__") or ext == ".recv":
                continue
            al = args.get(ext, [])
            al.append(filename)
            args[ext] = al
            pass
        for ext in args.keys():
            if ext == ".recv":
                continue
            dst = "__all.recv%s" % (ext,)
            print_status(".")
            os.system("mergecap -w %s %s" % (dst, " ".join(args[ext])))
            print_status(".")
1152 1153 1154 1155 1156 1157 1158
            if ext != ".0":
                if not os.path.exists(ext[1:]):
                    os.mkdir(ext[1:])
                    pass
                os.chdir(ext[1:])
                dst = os.path.join(os.path.pardir, dst)
                pass
1159
            os.system("tcptrace -y -G %s > .tcptrace.out" % (dst,))
1160 1161 1162
            if ext != ".0":
                os.chdir(os.path.pardir)
                pass
1163
            pass
1164
        print_status("\r")
1165 1166 1167 1168 1169 1170 1171
        os.chdir(os.path.pardir)
        lpc += 1
        pass

    if len(link2trace.keys()):
        print_status("\n")
        pass
1172

1173 1174
    return

Timothy Stack's avatar
 
Timothy Stack committed
1175 1176 1177 1178
##
# Performs the "sync" action, which synchronizes the logholes for a group of
# nodes.
#
1179 1180
# XXX This method is in dire need of a rewrite.
#
Timothy Stack's avatar
 
Timothy Stack committed
1181 1182
# @param args Action-specific command line arguments.
#
1183
def do_sync(args, mode="sync", must_be_active=1, as_root=False):
Timothy Stack's avatar
 
Timothy Stack committed
1184
    from emulabclient import RESPONSE_SUCCESS
1185
    global OSIDS, SERVER, DOT_LOGHOLES, JOB_PROGRESS_INDEX
Timothy Stack's avatar
 
Timothy Stack committed
1186 1187 1188
    
    retval = 0

1189 1190 1191
    logdir = os.path.join(EXPDIR_FMT % { "PID" : PID, "EID" : EID }, "logs")

    standard_logholes = True
1192
    snapshot_traces = False
1193
    post_process = True
1194 1195 1196
    extra_logholes = []

    try:
1197
        opts, node_list = getopt.getopt(args, "nl:r:Ps", [
1198 1199 1200 1201 1202 1203
            "no-standard", "local=", "remote=",
            ])
        for opt, val in opts:
            if opt in ("-n", "--no-standard"):
                standard_logholes = False
                pass
1204 1205 1206
            elif opt in ("-P", "--no-post"):
                post_process = False
                pass
1207 1208 1209 1210 1211 1212 1213 1214 1215 1216
            elif opt in ("-l", "--local"):
                logdir = os.path.abspath(val)
                if not os.path.isdir(logdir):
                    raise getopt.error(
                        'error: local directory does not exist - '
                        + logdir)
                pass
            elif opt in ("-r", "--remote"):
                extra_logholes.append(val)
                pass
1217 1218 1219
            elif opt in ("-s", "--snapshot"):
                snapshot_traces = True
                pass
1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236
            pass
        pass
    except getopt.error, e:
        print e.args[0]
        sync_usage()
        return 2

    if not standard_logholes:
        DOT_LOGHOLES = ""
        pass
    
    for path in extra_logholes:
        DOT_LOGHOLES += loghole_include(path)
        pass

    DOT_LOGHOLES += "- *\n"
    
Timothy Stack's avatar
 
Timothy Stack committed
1237 1238 1239
    init_server()
    
    state_method = getattr(SERVER, "experiment.state")
Timothy Stack's avatar
 
Timothy Stack committed
1240
    
Timothy Stack's avatar
 
Timothy Stack committed
1241
    os.chdir(logdir)
1242 1243
    
    if standard_logholes and os.path.exists(CLEANFILE):
1244 1245 1246 1247
        do_clean(["-f", "-e"])
        os.unlink(CLEANFILE)
        pass

1248 1249 1250
    dlhandle = tempfile.NamedTemporaryFile(prefix=".logholes")
    dlhandle.write(DOT_LOGHOLES)
    dlhandle.flush()
Timothy Stack's avatar
 
Timothy Stack committed
1251 1252 1253 1254 1255 1256

    state = state_method(PACKAGE_VERSION, { "proj" : PID, "exp" : EID })
    if state["code"] != RESPONSE_SUCCESS:
        sys.stderr.write("error: " + state["output"] + "\n")
        retval = 1
        pass
1257 1258 1259
    elif (state["value"] != "active" and
          state["value"] != "activating" and
          state["value"] != "modify_reswap"):
Timothy Stack's avatar
 
Timothy Stack committed
1260 1261 1262 1263 1264 1265 1266 1267
        if must_be_active:
            sys.stderr.write("error: %s/%s is not an active experiment!\n" %
                             (PID, EID))
            retval = 2
            pass
        else:
            retval = 0
            pass
Timothy Stack's avatar
 
Timothy Stack committed
1268 1269 1270
        pass
    else:
        info_method = getattr(SERVER, "experiment.info")
Timothy Stack's avatar
 
Timothy Stack committed
1271
        osinfo_method = getattr(SERVER, "osid.info")
1272
        ssh_method = getattr(SERVER, "node.sshdescription")
Timothy Stack's avatar
 
Timothy Stack committed
1273

1274
        # Get any bits of information we need from the RPC server.
Timothy Stack's avatar
 
Timothy Stack committed
1275 1276 1277 1278 1279
        physical = info_method(PACKAGE_VERSION, {
            "proj" : PID,
            "exp" : EID,
            "aspect" : "mapping",
            })
Timothy Stack's avatar
 
Timothy Stack committed
1280

Timothy Stack's avatar
 
Timothy Stack committed
1281
        traces = info_method(PACKAGE_VERSION, {
Timothy Stack's avatar
 
Timothy Stack committed
1282 1283
            "proj" : PID,
            "exp" : EID,
Timothy Stack's avatar
 
Timothy Stack committed
1284
            "aspect" : "traces",
Timothy Stack's avatar
 
Timothy Stack committed
1285
            })
1286 1287 1288 1289 1290

        ssh_data = ssh_method(PACKAGE_VERSION, {
            "proj" : PID,
            "exp" : EID
            })
Timothy Stack's avatar
 
Timothy Stack committed
1291
        
Timothy Stack's avatar
 
Timothy Stack committed
1292
        if ((physical["code"] != RESPONSE_SUCCESS) or
1293 1294
            (traces["code"] != RESPONSE_SUCCESS) or
            (ssh_data["code"] != RESPONSE_SUCCESS)):
Timothy Stack's avatar
 
Timothy Stack committed
1295 1296 1297 1298 1299 1300
            sys.stderr.write("error: xmlrpc failed - %s\n" %
                             (physical["output"],))
            retval = 1
            pass
        else:
            physical = physical["value"]
Timothy Stack's avatar
 
Timothy Stack committed
1301
            traces = traces["value"]
1302
            ssh_data = ssh_data["value"]
1303 1304 1305 1306

            if snapshot_traces:
                trace_pre(traces)
                pass
1307 1308

            # Separate the nodes into those that are up
Timothy Stack's avatar
 
Timothy Stack committed
1309
            hosts = [(x["node"], x["name"]) for x in physical.values()
Timothy Stack's avatar
 
Timothy Stack committed
1310 1311 1312
                     if ((x["erole"] == "virthost" or
                          x["erole"] == "node" or
                          x["erole"] == "delaynode") and
Timothy Stack's avatar
 
Timothy Stack committed
1313 1314
                         (x["status"] == "up" and x["eventstatus"] == "ISUP"))]

1315
            # ... or down.
1316 1317 1318 1319 1320 1321
            downhosts = [(x["node"], x["name"]) for x in physical.values()
                     if ((x["erole"] == "virthost" or
                          x["erole"] == "node" or
                          x["erole"] == "delaynode") and
                         (x["status"] != "up" or x["eventstatus"] != "ISUP"))]

Timothy Stack's avatar
 
Timothy Stack committed
1322 1323
            phosts = sets.Set([x[0] for x in hosts])
            vhosts = sets.Set([x[1] for x in hosts])
1324 1325
            dphosts = sets.Set([x[0] for x in downhosts])
            dvhosts = sets.Set([x[1] for x in downhosts])
1326 1327
            if len(node_list) == 0:
                node_list = vhosts
1328 1329 1330 1331 1332
                
                for host in downhosts:
                    sys.stderr.write("warning: '%s' is down, its logs will "
                                     "not be retrieved\n" % (host[1],))
                    pass
Timothy Stack's avatar
 
Timothy Stack committed
1333 1334
                pass

1335 1336
            known_nodes = sets.Set(node_list) & (phosts | vhosts)
            unknown_nodes = known_nodes ^ sets.Set(node_list)
Timothy Stack's avatar
 
Timothy Stack committed
1337

Timothy Stack's avatar
 
Timothy Stack committed
1338 1339
            jobs = {}
            hoststatus = {}
1340
            hoststart = {}
1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351
            hosttries = {}
            logsize = {}
            while len(hosts) > 0 or len(jobs) > 0:
                if len(hosts) > 0:
                    ht = hosts.pop()
                    phost, vhost = ht
                    pass
                else:
                    phost = ""
                    vhost = ""
                    pass
1352
                if (vhost in node_list) or (phost in node_list):
1353
                    # Start a job for the next host.
Timothy Stack's avatar
 
Timothy Stack committed
1354 1355 1356
                    osid = physical[vhost]["osid"]
                    if osid not in OSIDS:
                        oi = osinfo_method(PACKAGE_VERSION, {
1357
                            "osid" : osid })
Timothy Stack's avatar
 
Timothy Stack committed
1358 1359 1360 1361 1362
                        if oi["code"] == RESPONSE_SUCCESS:
                            OSIDS[osid] = oi["value"]
                            pass
                        else:
                            OSIDS[osid] = None
Timothy Stack's avatar
 
Timothy Stack committed
1363 1364
                            pass
                        pass
1365

1366
                    # Make sure we can talk to it.
Timothy Stack's avatar
 
Timothy Stack committed
1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381
                    if not OSIDS[osid]:
                        sys.stderr.write("warning: unknown OSID %s for %s\n" %
                                         (physical[vhost]["osid"], vhost))
                        continue
                    else:
                        vos = OSIDS[physical[vhost]["osid"]]
                        if "ssh" not in vos["osfeatures"]:
                            if VERBOSITY > VERBOSITY_HUMAN:
                                sys.stderr.write("warning: %s OS, %s, does "
                                                 "not support ssh, skipping\n"
                                                 % (vhost, vos["osname"]))
                                pass
                            continue
                        pass

1382 1383 1384 1385 1386 1387 1388
                    # Start the job.
                    if ht in hosttries:
                        delay = 3
                        pass
                    else:
                        delay = 0
                        pass
Timothy Stack's avatar
 
Timothy Stack committed
1389
                    if mode == "sync":
1390 1391
                        pid = rsync(phost, os.path.join(logdir, vhost),
                                    dlhandle.name,
1392
                                    port = ssh_data[phost].get("port", None),
1393
                                    output = True,
1394 1395
                                    delay = delay,
                                    as_root = as_root)
Timothy Stack's avatar
 
Timothy Stack committed
1396 1397
                        pass
                    elif mode == "clean":
1398 1399 1400
                        pid = rmstar(phost,
                                     ssh_data[phost].get("port", None),
                                     delay,
1401
                                     as_root,
Timothy Stack's avatar
Timothy Stack committed
1402
                                     "/local/logs/*")
Timothy Stack's avatar
 
Timothy Stack committed
1403 1404 1405 1406
                        pass
                    else:
                        assert 0
                    
1407 1408 1409
                    jobs[pid] = ht
                    logsize[pid] = 0
                    hoststart[ht] = time.time()
Timothy Stack's avatar
 
Timothy Stack committed
1410

1411 1412 1413 1414 1415 1416 1417 1418 1419 1420 1421 1422 1423 1424 1425 1426 1427 1428 1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440
                    if ht not in hosttries:
                        hosttries[ht] = 0
                        pass
                    hosttries[ht] += 1
                    pass
                
                try:
                    # Wait for the children and update the status.
                    while len(jobs) == PARALLELIZATION or len(hosts) == 0:
                        print_job_status(mode, node_list, jobs,
                                         hoststatus, hoststart)
                        # Check for any changes in the log files.
                        if (mode == "sync" and
                            making_progress(jobs, logsize, logdir)):
                            JOB_PROGRESS_INDEX += 1
                            pass
                        # Check for processes that have ended.
                        pid, sts = os.waitpid(0, os.WNOHANG)
                        if pid in jobs:
                            # Retry on some errors 
                            if (mode == "clean" and sts == 255 and
                                hosttries[jobs[pid]] <= JOB_TRIES):
                                hosts.append(jobs[pid])
                                pass
                            elif (mode == "sync" and
                                  sts not in NONRETRIABLE_RSYNC_STATUS and
                                  hosttries[jobs[pid]] <= JOB_TRIES):
                                hosts.append(jobs[pid])
                                pass
                            else:
1441
                                hoststatus[jobs[pid]] = (time.time(), sts)
Timothy Stack's avatar
 
Timothy Stack committed
1442
                                pass
1443
                            del jobs[pid]
Timothy Stack's avatar
 
Timothy Stack committed
1444
                            pass
Timothy Stack's avatar
 
Timothy Stack committed
1445

1446 1447 1448
                        if len(jobs) == PARALLELIZATION or len(hosts) == 0:
                            time.sleep(0.5)
                            pass
1449
                        pass
Timothy Stack's avatar
 
Timothy Stack committed
1450
                    pass
1451 1452
                except OSError:
                    pass
Timothy Stack's avatar
 
Timothy Stack committed
1453 1454
                pass

1455 1456
            sys.stdout.write("\n")

1457 1458
            # Clue the user in if there were nodes that they specified nodes
            # we don't know about.
Timothy Stack's avatar
 
Timothy Stack committed
1459 1460
            if len(unknown_nodes) > 0:
                for vhost in unknown_nodes:
1461 1462 1463 1464 1465 1466 1467 1468
                    if vhost in dvhosts or vhost in dphosts:
                        sys.stderr.write("error: node '%s' is down\n" %
                                         (vhost,))
                        pass
                    else:
                        sys.stderr.write("error: unknown node - %s\n" %
                                         (vhost,))
                        pass
Timothy Stack's avatar
 
Timothy Stack committed
1469 1470 1471
                    pass
                retval = 2
                pass
Timothy Stack's avatar
 
Timothy Stack committed
1472

1473 1474 1475
            # Check the status of the jobs and dump logs for any that failed.
            for ht, (etime, sts) in hoststatus.items():
                phost, vhost = ht
1476
                if VERBOSITY > VERBOSITY_HUMAN:
1477
                    print "info: %s - %.2f" % (vhost, etime - hoststart[ht])
1478
                    pass
1479 1480
                if mode == "clean":
                    continue
Timothy Stack's avatar
 
Timothy Stack committed
1481
                if sts not in ACCEPTABLE_RSYNC_STATUS:
Timothy Stack's avatar
 
Timothy Stack committed
1482 1483
                    sys.stderr.write("error: failed to %s node '%s'\n" %
                                     (mode, vhost))
1484 1485 1486 1487 1488
                    eout = open(os.path.join(logdir, vhost, ".rsync.out"),
                                "r").read()
                    sys.stderr.write(
                        vhost + ": " + re.sub(r'\n',
                                              "\n" + vhost + ": ",
1489
                                              eout[:-1]) + eout[-1])
Timothy Stack's avatar
 
Timothy Stack committed
1490 1491
                    retval = 3
                    pass
1492
                else:
1493 1494 1495 1496 1497
                    # Post process the delay-agent log files, if there are any.
                    vdir = os.path.join(vhost, "var", "emulab", "logs")
                    if (os.path.exists(vdir) and
                        os.path.exists("delayagent.debug")):
                        os.chdir(vdir)