[tor-commits] [tor/master] Allow fallbacks serving consensuses that expired less than 24 hours ago

[nickm@xxxxxxxxxxxxxx]

commit 654367f0260ff144898dc0ac2a45b710e7350f29
Author: teor <teor2345@xxxxxxxxx>
Date:   Sat Dec 10 21:35:23 2016 +1100

    Allow fallbacks serving consensuses that expired less than 24 hours ago
    
    This works around #20909, where relays serve stale consensuses for a short
    time, and then recover.
    
    Update to the fix for #20539.
---
 changes/fallbacks-201612            |  3 ++-
 scripts/maint/updateFallbackDirs.py | 35 +++++++++++++++++++++++++++--------
 2 files changed, 29 insertions(+), 9 deletions(-)

diff --git a/changes/fallbacks-201612 b/changes/fallbacks-201612
index 5bb7bdd..03d9843 100644
--- a/changes/fallbacks-201612
+++ b/changes/fallbacks-201612
@@ -13,7 +13,8 @@
     - Make it easier to change the output sort order of fallbacks.
       Closes ticket 20822.
     - Exclude relays affected by 20499 from the fallback list. Exclude known
-      affected versions, and any relay that delivers a stale consensus.
+      affected versions, and any relay that delivers a stale consensus, as
+      long as that consensus expired more than 24 hours ago.
       Closes ticket 20539.
     - Require fallbacks to have flags for 90% of the time (weighted decaying
       average), rather than 95%. This allows at least 73% of clients to
diff --git a/scripts/maint/updateFallbackDirs.py b/scripts/maint/updateFallbackDirs.py
index 41d210f..4f60fd4 100755
--- a/scripts/maint/updateFallbackDirs.py
+++ b/scripts/maint/updateFallbackDirs.py
@@ -88,6 +88,19 @@ MUST_BE_RUNNING_NOW = (PERFORM_IPV4_DIRPORT_CHECKS
 # Clients have been using microdesc consensuses by default for a while now
 DOWNLOAD_MICRODESC_CONSENSUS = True
 
+# If a relay delivers an expired consensus, if it expired less than this many
+# seconds ago, we still allow the relay. This should never be less than -90,
+# as all directory mirrors should have downloaded a consensus 90 minutes
+# before it expires. It should never be more than 24 hours, because clients
+# reject consensuses that are older than REASONABLY_LIVE_TIME.
+# For the consensus expiry check to be accurate, the machine running this
+# script needs an accurate clock.
+# We use 24 hours to compensate for #20909, where relays on 0.2.9.5-alpha and
+# 0.3.0.0-alpha-dev and later deliver stale consensuses, but typically recover
+# after ~12 hours.
+# We should make this lower when #20909 is fixed, see #20942.
+CONSENSUS_EXPIRY_TOLERANCE = 24*60*60
+
 # Output fallback name, flags, bandwidth, and ContactInfo in a C comment?
 OUTPUT_COMMENTS = True if OUTPUT_CANDIDATES else False
 
@@ -1146,7 +1159,6 @@ class Candidate(object):
   def fallback_consensus_download_speed(dirip, dirport, nickname, fingerprint,
                                         max_time):
     download_failed = False
-    start = datetime.datetime.utcnow()
     # some directory mirrors respond to requests in ways that hang python
     # sockets, which is why we log this line here
     logging.info('Initiating %sconsensus download from %s (%s:%d) %s.',
@@ -1155,6 +1167,7 @@ class Candidate(object):
     # there appears to be about 1 second of overhead when comparing stem's
     # internal trace time and the elapsed time calculated here
     TIMEOUT_SLOP = 1.0
+    start = datetime.datetime.utcnow()
     try:
       consensus = get_consensus(
                               endpoints = [(dirip, dirport)],
@@ -1165,26 +1178,32 @@ class Candidate(object):
                               document_handler = DocumentHandler.BARE_DOCUMENT,
                               microdescriptor = DOWNLOAD_MICRODESC_CONSENSUS
                                 ).run()[0]
+      end = datetime.datetime.utcnow()
+      time_since_expiry = (end - consensus.valid_until).total_seconds()
     except Exception, stem_error:
+      end = datetime.datetime.utcnow()
       logging.info('Unable to retrieve a consensus from %s: %s', nickname,
                     stem_error)
       status = 'error: "%s"' % (stem_error)
       level = logging.WARNING
       download_failed = True
-    elapsed = (datetime.datetime.utcnow() - start).total_seconds()
+    elapsed = (end - start).total_seconds()
     if download_failed:
-      # keep the error failure status
+      # keep the error failure status, and avoid using the variables
       pass
     elif elapsed > max_time:
       status = 'too slow'
       level = logging.WARNING
       download_failed = True
-    elif datetime.datetime.utcnow() > consensus.valid_until:
-      time_since_expiry = (datetime.datetime.utcnow() -
-                           consensus.valid_until).total_seconds()
+    elif (time_since_expiry > 0):
       status = 'outdated consensus, expired %ds ago'%(int(time_since_expiry))
-      level = logging.WARNING
-      download_failed = True
+      if time_since_expiry <= CONSENSUS_EXPIRY_TOLERANCE:
+        status += ', tolerating up to %ds'%(CONSENSUS_EXPIRY_TOLERANCE)
+        level = logging.INFO
+      else:
+        status += ', invalid'
+        level = logging.WARNING
+        download_failed = True
     else:
       status = 'ok'
       level = logging.DEBUG



_______________________________________________
tor-commits mailing list
tor-commits@xxxxxxxxxxxxxxxxxxxx
https://lists.torproject.org/cgi-bin/mailman/listinfo/tor-commits