autobuild: timeout and reporting cleanup
v3 cover autobuild: timeout and reporting cleanup
Included in this series
- Timeout detection using modification time of build-time.log
- Cleanup build reason reporting to use build-time.log last entry
- (Misc) Noticed schema was missing the new branch field
Related items needing merge
- Force sane umask to ensure expected log output
(https://patchwork.ozlabs.org/patch/857788/)
Matt Weber (3):
autobuilder: hung build: convert to monitor thread
autobuilder: schema update for branch varchar
autobuilder: failure reason use build-time.log
scripts/autobuild-run | 58 +++++++++++++++++++++++++++++++++++++++++++-------- web/import.inc.php | 7 ++++++- web/schema.sql | 1 + 3 files changed, 56 insertions(+), 10 deletions(-)
Signature · 1 line
2.14.2
PATCH v3 1/1 autobuilder: failure reason use build-time.log
When the make output doesn't parse to provide a result and unknown is assumed, check the build-time.log and see if the last line was a <stamp>:start:<pkgname>. IF so, use this pkgname instead before declaring unknown.
This was tested against the ti-cgt-pru hang.
http://autobuild.buildroot.net/results/60e/60e11a3bb90b9f41259e4a970081b72d8b8d100b//
--- web/import.inc.php | 7 ++++++- 1 file changed, 6 insertions(+), 1 deletion(-)
web/import.inc.php +6 -1
@@ -235,7 +235,12 @@ function import_result($buildid, $filename) if (trim($tmp[0])) $reason = $tmp[0]; else - $reason = "unknown"; + exec("tail -1 " . $thisbuildfinaldir . "build-time.log | grep :start: | cut -d':' -f4", $tmp); + if (trim($tmp[0])) + print "Using build-time.log for reason[".trim($tmp[0])."]"; + $reason = trim($tmp[0]); + else + $reason = "unknown"; } $db = new db();
Signature · 1 line
2.14.2
PATCH v3 1/3 autobuilder: hung build: convert to monitor thread
In reply to Matt WeberCheck the build-time.log and monitor for modifications to determine if the build has hung for at most #mins before killing the build and reporting a timeout.
This allows infinite sized builds as we get to a lower number of autobr fails. Less failures means we start to see false timeout failures when we hit the boundary of the old MAX_DURATION ~8hrs.
--
Change Log
v2 -> v3
- Adjust hung timeout to 2hrs as minic, gst-ffmpeg and qt5webkit could go beyond 60mins on a minimal 2-4core machine while processing a single build step written
v1->v2
[Thomas P.
- Use mtime vs reading file
- Use datetime for hung delta check
- Removed camel case
- Added hung build event to sync hand-off back to main thread
--- scripts/autobuild-run | 58 +++++++++++++++++++++++++++++++++++++++++++-------- 1 file changed, 49 insertions(+), 9 deletions(-)
scripts/autobuild-run +49 -9
@@ -143,6 +143,8 @@ import sys from time import localtime, strftime from distutils.version import StrictVersion import platform +from threading import Thread, Event +import datetime if sys.hexversion >= 0x3000000: import configparser
@@ -167,7 +169,12 @@ else: decode_bytes = _identity encode_str = _identity -MAX_DURATION = 60 * 60 * 8 +# The following pkgs can be > 60mins of build time +# gst-ffmpeg - http://autobuild.buildroot.net/results/5f7/5f7d1847ebd65f221bf18095decaa3383d24a89c/ +# qt5webkit - http://autobuild.buildroot.net/results/195/195dc9ad4b21f6e7675bed277209ba2480337d54/ +# mimic - http://autobuild.buildroot.net/results/ae6/ae66d86988d8c7c0ae19597fed9dee4fafd48f90/ +# +HUNG_BUILD_TIMEOUT = 120 # mins VERSION = 1 def log_write(logf, msg):
@@ -199,7 +206,7 @@ def get_branch(): return branches[randint(0, len(branches) - 1)] class SystemInfo: - DEFAULT_NEEDED_PROGS = ["make", "git", "gcc", "timeout"] + DEFAULT_NEEDED_PROGS = ["make", "git", "gcc"] DEFAULT_OPTIONAL_PROGS = ["bzr", "java", "javac", "jar"] def __init__(self):
@@ -358,6 +365,24 @@ def gen_config(**kwargs): ret = subprocess.call(args, stdout=devnull, stderr=log) return ret +def stop_on_build_hang(monitor_thread_hung_build_flag, + monitor_thread_stop_flag, + sub_proc, outputdir, log): + build_time_logfile = os.path.join(outputdir, "build/build-time.log") + while True: + if monitor_thread_stop_flag.is_set(): + return + if os.path.exists(build_time_logfile): + mtime = datetime.datetime.fromtimestamp(os.stat(build_time_logfile).st_mtime) + + if mtime < datetime.datetime.now() - datetime.timedelta(minutes=HUNG_BUILD_TIMEOUT): + if sub_proc.poll() is None: + monitor_thread_hung_build_flag.set() # Used by do_build() to determine build hang + log_write(log, "INFO: build hung") + sub_proc.kill() + break + monitor_thread_stop_flag.wait(30) + def do_build(**kwargs): """Run the build itself"""
@@ -375,25 +400,40 @@ def do_build(**kwargs): f = open(os.path.join(outputdir, "logfile"), "w+") log_write(log, "INFO: build started") - cmd = ["timeout", str(MAX_DURATION), - "nice", "-n", str(nice), + cmd = ["nice", "-n", str(nice), "make", "O=%s" % outputdir, "-C", srcdir, "BR2_DL_DIR=%s" % dldir, "BR2_JLEVEL=%s" % kwargs['njobs']] \ + kwargs['make_opts'].split() sub = subprocess.Popen(cmd, stdout=f, stderr=f) + + # Setup hung build monitoring thread + monitor_thread_hung_build_flag = Event() + monitor_thread_stop_flag = Event() + build_monitor = Thread(target=stop_on_build_hang, + args=(monitor_thread_hung_build_flag, + monitor_thread_stop_flag, + sub, outputdir, log)) + build_monitor.daemon = True + build_monitor.start() + kwargs['buildpid'][kwargs['instance']] = sub.pid ret = sub.wait() kwargs['buildpid'][kwargs['instance']] = 0 - # 124 is a special error code that indicates we have reached the - # timeout - if ret == 124: - log_write(log, "INFO: build timed out") + # If build failed, monitor thread would have exited@this point + if monitor_thread_hung_build_flag.is_set(): + log_write(log, "INFO: build timed out [%d]" % ret) return -2 + else: + # Stop monitor thread as this build didn't timeout + monitor_thread_stop_flag.set() + # Monitor thread should be exiting around this point + if ret != 0: - log_write(log, "INFO: build failed") + log_write(log, "INFO: build failed [%d]" % ret) return -1 + cmd = ["make", "O=%s" % outputdir, "-C", srcdir, "BR2_DL_DIR=%s" % dldir, "legal-info"] \ + kwargs['make_opts'].split()
Signature · 1 line
2.14.2
PATCH v3 1/3 autobuilder: hung build: convert to monitor thread
Hello,
Quoted 25 lines · Matt Weber
Check the build-time.log and monitor for modifications to determine if the build has hung for at most #mins before killing the build and reporting a timeout.
This allows infinite sized builds as we get to a lower number of autobr fails. Less failures means we start to see false timeout failures when we hit the boundary of the old
MAX_DURATION~8hrs.Signed-off-by Matthew Weber <[email protected]>--
Change Logv2 -> v3
- Adjust hung timeout to 2hrs as minic, gst-ffmpeg and qt5webkit could go beyond 60mins on a minimal 2-4core machine while processing a single build step written
v1->v2
[Thomas P.
- Use mtime vs reading file
- Use datetime for hung delta check
- Removed camel case
- Added hung build event to sync hand-off back to main thread
Quoted 3 lines
--- scripts/autobuild-run | 58 +++++++++++++++++++++++++++++++++++++++++++-------- 1 file changed, 49 insertions(+), 9 deletions(-)
Applied to buildroot-test, thanks!. Note that in this patch, and other patches in this series, your From: didn't match the Signed-off-by, so I had to rewrite these.
Thomas
Signature · 3 lines
Thomas Petazzoni, CTO, Bootlin (formerly Free Electrons) Embedded Linux and Kernel engineering https://bootlin.com
PATCH v3 2/3 autobuilder: schema update for branch varchar
In reply to Matt Weber--- web/schema.sql | 1 + 1 file changed, 1 insertion(+)
web/schema.sql +1 -0
@@ -11,6 +11,7 @@ CREATE TABLE `results` ( `static` tinyint(1) NOT NULL default '0', `subarch` varchar(64) NOT NULL DEFAULT '', `duration` int(11) NOT NULL DEFAULT '0', + `branch` varchar(64) NOT NULL DEFAULT '', PRIMARY KEY (`id`) ) ENGINE=InnoDB DEFAULT CHARSET=latin1;
Signature · 1 line
2.14.2
PATCH v3 2/3 autobuilder: schema update for branch varchar
Hello,
Quoted 1 line · Matt Weber
Signed-off-by Matthew Weber <[email protected]>
Applied to buildroot-test.
Thomas
Signature · 3 lines
Thomas Petazzoni, CTO, Bootlin (formerly Free Electrons) Embedded Linux and Kernel engineering http://bootlin.com
PATCH v3 3/3 autobuilder: failure reason use build-time.log
In reply to Matt WeberWhen the make output doesn't parse to provide a result and unknown is assumed, check the build-time.log and see if the last line was a <stamp>:start:<pkgname>. IF so, use this pkgname instead before declaring unknown.
This was tested against the ti-cgt-pru hang.
http://autobuild.buildroot.net/results/60e/60e11a3bb90b9f41259e4a970081b72d8b8d100b//
--- web/import.inc.php | 7 ++++++- 1 file changed, 6 insertions(+), 1 deletion(-)
web/import.inc.php +6 -1
@@ -235,7 +235,12 @@ function import_result($buildid, $filename) if (trim($tmp[0])) $reason = $tmp[0]; else - $reason = "unknown"; + exec("tail -1 " . $thisbuildfinaldir . "build-time.log | grep :start: | cut -d':' -f4", $tmp); + if (trim($tmp[0])) + print "Using build-time.log for reason[".trim($tmp[0])."]"; + $reason = trim($tmp[0]); + else + $reason = "unknown"; } $db = new db();
Signature · 1 line
2.14.2
PATCH v3 3/3 autobuilder: failure reason use build-time.log
Hello,
web/import.inc.php function import_result($buildid, $filename) · line 241
Quoted 4 lines · Matt Weber
web/import.inc.php +4
-0+ exec("tail -1 " . $thisbuildfinaldir . "build-time.log | grep :start: | cut -d':' -f4", $tmp); + if (trim($tmp[0])) + print "Using build-time.log for reason[".trim($tmp[0])."]"; + $reason = trim($tmp[0]);
Curly braces missing here. I've added a bunch more curly braces in this piece of code, and applied to buildroot-test.
Thanks,
Thomas
Signature · 3 lines
Thomas Petazzoni, CTO, Bootlin (formerly Free Electrons) Embedded Linux and Kernel engineering http://bootlin.com