By delaying the agent key query until after the fork, we prevent the
problem of simultaneous access to the agent.
Tested that it works against 80 hosts in parallel without error; the
current version breaks already at 20 hosts.
---
tools/ganeti-listrunner | 25 ++++++++++++++-----------
1 files changed, 14 insertions(+), 11 deletions(-)
diff --git a/tools/ganeti-listrunner b/tools/ganeti-listrunner
index fba0b75..1f63c1d 100755
--- a/tools/ganeti-listrunner
+++ b/tools/ganeti-listrunner
@@ -1,7 +1,7 @@
#!/usr/bin/python
#
-# Copyright (C) 2006, 2007, 2010 Google Inc.
+# Copyright (C) 2006, 2007, 2010, 2011 Google Inc.
#
# This program is free software; you can redistribute it and/or modify
# it under the terms of the GNU General Public License as published by
@@ -203,7 +203,7 @@ def GetAgentKeys():
return []
-def SetupSshConnection(host, username, password, keys, logfile):
+def SetupSshConnection(host, username, password, use_agent, logfile):
"""Setup the ssh connection used for all later steps.
This function sets up the ssh connection that will be used both
@@ -219,6 +219,10 @@ def SetupSshConnection(host, username, password, keys,
logfile):
print " - ERROR: host not reachable on 22/tcp"
return False
+ if use_agent:
+ keys = GetAgentKeys()
+ else:
+ keys = []
all_kwargs = [{"pkey": k} for k in keys]
all_desc = ["key %d" % d for d in range(len(keys))]
if password is not None:
@@ -346,7 +350,7 @@ def RunRemoteCommand(connection, command, logfile):
return True
-def HostWorker(logdir, username, password, keys, hostname,
+def HostWorker(logdir, username, password, use_agent, hostname,
executable, command, filelist):
"""Per-host worker.
@@ -357,7 +361,7 @@ def HostWorker(logdir, username, password, keys, hostname,
@param logdir: the directory where the logfiles must be created
@param username: SSH username
@param password: SSH password
- @param keys: SSH keys
+ @param use_agent: whether we should instead use an agent
@param hostname: the hostname to connect to
@param executable: the executable to upload, if not None
@param command: the command to run
@@ -370,7 +374,7 @@ def HostWorker(logdir, username, password, keys, hostname,
result = 0 # optimism, I know
try:
connection = SetupSshConnection(hostname, username,
- password, keys, logfile)
+ password, use_agent, logfile)
if connection is not False:
if executable is not None:
print " %s: uploading files" % hostname
@@ -412,7 +416,7 @@ def HostWorker(logdir, username, password, keys, hostname,
sys.exit(result)
-def LaunchWorker(child_pids, logdir, username, password, keys, hostname,
+def LaunchWorker(child_pids, logdir, username, password, use_agent, hostname,
executable, command, filelist):
"""Launch the per-host worker.
@@ -426,7 +430,7 @@ def LaunchWorker(child_pids, logdir, username, password,
keys, hostname,
# controller just record the pids
child_pids[pid] = hostname
else:
- HostWorker(logdir, username, password, keys, hostname,
+ HostWorker(logdir, username, password, use_agent, hostname,
executable, command, filelist)
@@ -489,9 +493,8 @@ def main():
print "ERROR: cannot create logfiles in dir %s, aborting" % logdir
sys.exit(1)
- keys = []
if use_agent:
- keys = GetAgentKeys()
+ pass
elif password:
try:
fh = file(password)
@@ -523,7 +526,7 @@ def main():
hosts = hosts[batch_size:]
child_pids = {}
for hostname in batch:
- LaunchWorker(child_pids, logdir, username, password, keys, hostname,
+ LaunchWorker(child_pids, logdir, username, password, use_agent, hostname,
executable, command, filelist)
while child_pids:
@@ -535,7 +538,7 @@ def main():
else:
failures += 1
if hosts:
- LaunchWorker(child_pids, logdir, username, password, keys,
+ LaunchWorker(child_pids, logdir, username, password, use_agent,
hosts.pop(0), executable, command, filelist)
print
--
1.7.3.1
