Reduce timeout conditions for Chrome OS for efficiency.
Many of the Autotest timeouts are far too long for Chrome OS. Where a
normal machine might take 10+ minutes to recover, Chromebooks take only
seconds. As such, I've tweaked many of the boot and crash timeouts to
reflect this.
When a host goes down the scheduler idles the job for four hours while
waiting for crash collection, causing two main problems:
1. Metahost jobs take four hours to requeue upon verify failure.
2. Results are delayed four hours when a host dies in jobs.
Additional issues fixed:
1. Timeout after reboot was essentially doubled in the default Host
implementation of wait_for_reboot. Works for us, but will need to query
upstream to ensure it works for everyone.
2. Unified timeouts across autotest.py and base_classes.py so boot
timeouts exist in only one place. Should be okay since all hosts extend
from Host in base_classes.py. Timings are now configurable in
global_config.
3. Global SSH options to ensure better timings. In order to do this, it
was necessary to add site extensibility to abstract_ssh.
Changes to crashcollect.py, autotest.py, abstract_ssh.py, and
base_classes.py will be upstreamed.
BUG=none
TEST=run_remote_tests. Will test in production as well.
Change-Id: I92964b140337e4ab919288b5190bb2e6d5a4fa2d
Reviewed-on: http://gerrit.chromium.org/gerrit/2269
Reviewed-by: Paul Pendlebury <pauldean@chromium.org>
Reviewed-by: Scott Zawalski <scottz@chromium.org>
Tested-by: Dale Curtis <dalecurtis@chromium.org>
diff --git a/server/autotest.py b/server/autotest.py
index 8591efd..726d043 100644
--- a/server/autotest.py
+++ b/server/autotest.py
@@ -7,14 +7,9 @@
from autotest_lib.client.common_lib import global_config, packages
from autotest_lib.client.common_lib import utils as client_utils
-AUTOTEST_SVN = 'svn://test.kernel.org/autotest/trunk/client'
+AUTOTEST_SVN = 'svn://test.kernel.org/autotest/trunk/client'
AUTOTEST_HTTP = 'http://test.kernel.org/svn/autotest/trunk/client'
-# Timeouts for powering down and up respectively
-HALT_TIME = 300
-BOOT_TIME = 1800
-CRASH_RECOVERY_TIME = 9000
-
get_value = global_config.global_config.get_config_value
autoserv_prebuild = get_value('AUTOSERV', 'enable_server_prebuild',
@@ -37,7 +32,7 @@
implement the unimplemented methods in parent classes.
"""
- def __init__(self, host = None):
+ def __init__(self, host=None):
self.host = host
self.got = False
self.installed = False
@@ -223,7 +218,7 @@
except (error.PackageInstallError, error.AutoservRunError,
global_config.ConfigError), e:
logging.info("Could not install autotest using the packaging "
- "system: %s. Trying other methods", e)
+ "system: %s. Trying other methods", e)
# try to install from file or directory
if self.source_material:
@@ -272,7 +267,7 @@
self.installed = False
- def get(self, location = None):
+ def get(self, location=None):
if not location:
location = os.path.join(self.serverdir, '../client')
location = os.path.abspath(location)
@@ -290,7 +285,7 @@
def run(self, control_file, results_dir='.', host=None, timeout=None,
tag=None, parallel_flag=False, background=False,
- client_disconnect_timeout=1800):
+ client_disconnect_timeout=None):
"""
Run an autotest job on the remote machine.
@@ -307,7 +302,8 @@
a background job; the code calling run will be responsible
for monitoring the client and collecting the results.
@param client_disconnect_timeout: Seconds to wait for the remote host
- to come back after a reboot. [default: 30 minutes]
+ to come back after a reboot. Defaults to the host setting for
+ DEFAULT_REBOOT_TIMEOUT.
@raises AutotestRunError: If there is a problem executing
the control file.
@@ -315,6 +311,9 @@
host = self._get_host_and_setup(host)
results_dir = os.path.abspath(results_dir)
+ if client_disconnect_timeout is None:
+ client_disconnect_timeout = host.DEFAULT_REBOOT_TIMEOUT
+
if tag:
results_dir = os.path.join(results_dir, tag)
@@ -700,12 +699,13 @@
def _wait_for_reboot(self, old_boot_id):
logging.info("Client is rebooting")
logging.info("Waiting for client to halt")
- if not self.host.wait_down(HALT_TIME, old_boot_id=old_boot_id):
+ if not self.host.wait_down(self.host.WAIT_DOWN_REBOOT_TIMEOUT,
+ old_boot_id=old_boot_id):
err = "%s failed to shutdown after %d"
- err %= (self.host.hostname, HALT_TIME)
+ err %= (self.host.hostname, self.host.WAIT_DOWN_REBOOT_TIMEOUT)
raise error.AutotestRunError(err)
logging.info("Client down, waiting for restart")
- if not self.host.wait_up(BOOT_TIME):
+ if not self.host.wait_up(self.host.DEFAULT_REBOOT_TIMEOUT):
# since reboot failed
# hardreset the machine once if possible
# before failing this control file
@@ -719,7 +719,8 @@
warning %= self.host.hostname
logging.warning(warning)
raise error.AutotestRunError("%s failed to boot after %ds" %
- (self.host.hostname, BOOT_TIME))
+ (self.host.hostname,
+ self.host.DEFAULT_REBOOT_TIMEOUT))
self.host.reboot_followup()
@@ -765,7 +766,7 @@
self.log_unexpected_abort(logger)
# give the client machine a chance to recover from a crash
- self.host.wait_up(CRASH_RECOVERY_TIME)
+ self.host.wait_up(self.host.HOURS_TO_WAIT_FOR_RECOVERY * 3600)
msg = ("Aborting - unexpected final status message from "
"client on %s: %s\n") % (self.host.hostname, last)
raise error.AutotestRunError(msg)
diff --git a/server/crashcollect.py b/server/crashcollect.py
index 6e1de7f..dc582ac 100644
--- a/server/crashcollect.py
+++ b/server/crashcollect.py
@@ -1,5 +1,6 @@
import os, time, pickle, logging, shutil
+from autotest_lib.client.common_lib import global_config
from autotest_lib.server import utils
@@ -33,7 +34,12 @@
collect_uncollected_logs(host)
-def wait_for_machine_to_recover(host, hours_to_wait=4.0):
+# Load default for number of hours to wait before giving up on crash collection.
+HOURS_TO_WAIT = global_config.global_config.get_config_value(
+ 'SERVER', 'crash_collection_hours_to_wait', type=float, default=4.0)
+
+
+def wait_for_machine_to_recover(host, hours_to_wait=HOURS_TO_WAIT):
"""Wait for a machine (possibly down) to become accessible again.
@param host: A RemoteHost instance to wait on
diff --git a/server/hosts/abstract_ssh.py b/server/hosts/abstract_ssh.py
index 3723c46..0f61391 100644
--- a/server/hosts/abstract_ssh.py
+++ b/server/hosts/abstract_ssh.py
@@ -10,8 +10,8 @@
default=False)
-def make_ssh_command(user="root", port=22, opts='', hosts_file='/dev/null',
- connect_timeout=30, alive_interval=300):
+def _make_ssh_cmd_default(user="root", port=22, opts='', hosts_file='/dev/null',
+ connect_timeout=30, alive_interval=300):
base_command = ("/usr/bin/ssh -a -x %s -o StrictHostKeyChecking=no "
"-o UserKnownHostsFile=%s -o BatchMode=yes "
"-o ConnectTimeout=%d -o ServerAliveInterval=%d "
@@ -22,6 +22,11 @@
alive_interval, user, port)
+make_ssh_command = utils.import_site_function(
+ __file__, "autotest_lib.server.hosts.site_host", "make_ssh_command",
+ _make_ssh_cmd_default)
+
+
# import site specific Host class
SiteHost = utils.import_site_class(
__file__, "autotest_lib.server.hosts.site_host", "SiteHost",
diff --git a/server/hosts/site_host.py b/server/hosts/site_host.py
new file mode 100644
index 0000000..afe6342
--- /dev/null
+++ b/server/hosts/site_host.py
@@ -0,0 +1,22 @@
+def make_ssh_command(user='root', port=22, opts='', hosts_file=None,
+ connect_timeout=None, alive_interval=None):
+ """Override default make_ssh_command to use options tuned for Chrome OS.
+
+ Tuning changes:
+ - ConnectTimeout=10; maximum of 10 seconds allowed for an SSH connection
+ failure.
+
+ - ServerAliveInterval=60; which causes SSH to ping connection every
+ 60 seconds. In conjunction with ServerAliveCountMax ensures that if the
+ connection dies, Autotest will bail out quickly.
+
+ - ServerAliveCountMax=1; only allow a single keep alive failure.
+
+ - UserKnownHostsFile=/dev/null; we don't care about the keys. Host keys
+ change with every new installation, don't waste memory/space saving them.
+ """
+ base_command = ('/usr/bin/ssh -a -x %s -o StrictHostKeyChecking=no'
+ ' -o UserKnownHostsFile=/dev/null -o BatchMode=yes'
+ ' -o ConnectTimeout=10 -o ServerAliveInterval=60'
+ ' -o ServerAliveCountMax=1 -l %s -p %d')
+ return base_command % (opts, user, port)