Skip to contentSkip to conversation

autobuild-run improvements

PATCH Superseded Thomas De Schampheleire 4 participants 31 messages Dec 12, 2014 – Mar 18, 2015

lore
Share & tools
lore
#0 lore raw

v5 cover autobuild-run improvements

From: Thomas De Schampheleire <[email protected]>

This series against the buildroot-test repo makes some improvements to the autobuild-run script.

v5:
- add two patches to handle kill/interrupt properly
- make the kwargs dictionary passing to Process more readable
- rebase

v4:
- rename do_send_results into upload (Peter)

v3:
- introduce missing usage of do_send_results (and move patch backwards)
- update TODO list

v2:
- reorder patches
- take into account comments on boolean logic in first patch
- place docopt directly in scripts/ instead of in a subdirectory
- remove patch adding --git (Yann, Thomas)
- some bug fixes in the first version of the patches

As you can see, this iteration still has the docopt patches. Thomas Petazzoni has expressed his reservations against the introduction of docopt, because the benefits wouldn't weigh out against the addition of an external Python module. Thomas told me on IRC he would have a deeper look onto this before making a final decision.

In case the final decision is not to introduce docopt, following improvements could be done to the current code:
- save the arguments from the command-line into a dictionary instead of plain variables
- keep the added ini_config() method to also save the options in a dictionary, and use the added merge() method to merge both arguments and config options into one final configuration, without needing to handle it manually.

Thomas De Schampheleire (11):
scripts: add python module docopt
autobuild-run: use docopt for argument parsing
autobuild-run: add option --make-opts for custom Buildroot options
autobuild-run: use **kwargs to avoid explicit parameter passthroughs
autobuild-run: check-requirements does not need to know the login
details
autobuild-run: set LC_ALL=C to not use locale settings of host machine
autobuild-run: improve the logic to generate build-end.log
autobuild-run: save config.log files for failed package
autobuild-run: extend TODO list
autobuild-run: kill all children on SIGTERM
autobuild-run: catch KeyboardInterrupt in the same way as SIGTERM

 scripts/autobuild-run      | 350 +++++++++++++++++++--------
 scripts/docopt.LICENSE-MIT |  23 ++
 scripts/docopt.py          | 581 +++++++++++++++++++++++++++++++++++++++++++++
 3 files changed, 851 insertions(+), 103 deletions(-)
 create mode 100644 scripts/docopt.LICENSE-MIT
 create mode 100644 scripts/docopt.py
Signature · 1 line
1.8.5.1
#1 lore raw

PATCH v5 1/11 scripts: add python module docopt

From: Thomas De Schampheleire <[email protected]>

This commit adds the 'docopt' python module, version 0.6.1. Original source is: https://github.com/docopt/docopt
Since docopt stands on its own, only the .py file and the license is added, other files from github are skipped.

---
 scripts/docopt.LICENSE-MIT |  23 ++
 scripts/docopt.py          | 581 +++++++++++++++++++++++++++++++++++++++++++++
 2 files changed, 604 insertions(+)
 create mode 100644 scripts/docopt.LICENSE-MIT
 create mode 100644 scripts/docopt.py
scripts/docopt.LICENSE-MIT new file +23 -0
new file mode 100644
index 0000000..58ff1bc
@@ -0,0 +1,23 @@
+Copyright (c) 2012 Vladimir Keleshev, <[email protected]>
+
+Permission is hereby granted, free of charge, to any person
+obtaining a copy of this software and associated
+documentation files (the "Software"), to deal in the Software
+without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense,
+and/or sell copies of the Software, and to permit persons to
+whom the Software is furnished to do so, subject to the
+following conditions:
+
+The above copyright notice and this permission notice shall
+be included in all copies or substantial portions of the
+Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY
+KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE
+WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR
+PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR
+COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
scripts/docopt.py new file +581 -0

581 lines not on this page · raw message

Signature · 1 line
1.8.5.1
#2 lore raw

PATCH v5 2/11 autobuild-run: use docopt for argument parsing

In reply to Thomas De Schampheleire

From: Thomas De Schampheleire <[email protected]>

Using docopt, argument parsing becomes trivial. Adding a new argument is a matter of updating the usage text.
This commit removes the original cumbersome argument handling and uses docopt instead. A method is added to import the settings from the configuration file in a similar dictionary as the one created by docopt, so that both can be merged (giving priority to the configuration file, as before).

With these changes, each option can be passed as argument and in the configuration file. This means that http-user and http-password can now also be added as argument (even though passing the password on the command-line is not recommended).

---
 scripts/autobuild-run | 134 +++++++++++++++++++++++++++-----------------------
 1 file changed, 72 insertions(+), 62 deletions(-)
scripts/autobuild-run +72 -62
index 4d9ef90..b686a78 100755
@@ -57,6 +57,37 @@
 #   BR2_PACKAGE_CLASSPATH=y, improve the script to detect whether the
 #   necessary host machine requirements are there to build classpath.
 
+"""autobuild-run - run Buildroot autobuilder
+
+Usage: autobuild-run [options]
+
+Options:
+  -h, --help                     show this help message and exit
+  -V, --version                  show version
+  -n, --ninstances NINSTANCES    number of parallel instances [default: 1]
+  -j, --njobs NJOBS              number of parallel jobs [default: 1]
+  -s, --submitter SUBMITTER      name/machine of submitter [default: N/A]
+  --http-login LOGIN             username to send results with
+  --http-password PASSWORD       password to send results with (for security
+                                 reasons it is recommended to define this in the
+                                 config file instead, with user-read permissions
+                                 only)
+  -c, --config CONFIG            path to configuration file
+
+Format of the configuration file:
+
+  All arguments can also be specified in the configuration file specified with
+  --config, using 'key = value' format (not including the leading --
+  characters). For example:
+
+   [main]
+   ninstances = <value>
+   njobs = <value>
+   http-login = <value>
+   http-password = <value>
+   submitter = <value>
+"""
+
 import urllib2
 import csv
 from random import randint
@@ -70,6 +101,7 @@ import sys
 import hashlib
 import argparse
 import ConfigParser
+import docopt
 
 MAX_DURATION = 60 * 60 * 4
 VERSION = 1
@@ -522,81 +554,59 @@ def run_instance(instance, njobs, http_login, http_password, submitter, sysinfo)
         ret = do_build(instance, njobs, instance_log)
         send_results(instance, http_login, http_password, submitter, instance_log, ret)
 
-def config_get():
-    """Get configuration parameters, either from the command line or the config file"""
+# args / config file merging inspired by:
+# https://github.com/docopt/docopt/blob/master/examples/config_file_example.py
 
-    epilog_text = """
-Format of the configuration file:
+def load_ini_config(configfile):
+    """Load configuration from file, returning a docopt-like dictionary"""
 
-   [main]
-   ninstances = <value>
-   njobs = <value>
-   http-login = <value>
-   http-password = <value>
-   submitter = <value>
-"""
+    if not os.path.exists(configfile):
+        print "ERROR: configuration file %s does not exist" % configfile
+        sys.exit(1)
+
+    config = ConfigParser.RawConfigParser()
+    if not config.read(configfile):
+        print "ERROR: cannot parse configuration file %s" % configfile
+        sys.exit(1)
+
+    # Prepend '--' to options specified in the config file, so they can be
+    # merged with those given on the command-line
+    return dict(('--%s' % key, value) for key, value in config.items('main'))
 
-    parser = argparse.ArgumentParser(description='Run Buildroot autobuilder',
-                                     formatter_class=argparse.RawDescriptionHelpFormatter,
-                                     epilog=epilog_text)
-    parser.add_argument("--ninstances", '-n', metavar="NINSTANCES",
-                        help="Number of parallel instances", default=None)
-    parser.add_argument("--njobs", '-j', metavar="NJOBS",
-                        help="Number of parallel jobs", default=None)
-    parser.add_argument("--submitter", '-s', metavar="SUBMITTER",
-                        help="Name/machine of submitter")
-    parser.add_argument("--config", '-c', metavar="CONFIG",
-                        help="Path to configuration file")
-    args = parser.parse_args()
-
-    ninstances = 1
-    njobs = 1
-    http_login = None
-    http_password = None
-    submitter = "N/A"
-
-    if args.config:
-        if not os.path.exists(args.config):
-            print "ERROR: configuration file %s does not exist" % args.config
-            sys.exit(1)
-        parser = ConfigParser.RawConfigParser()
-        if not parser.read([args.config]):
-            print "ERROR: cannot parse configuration file %s" % args.config
-            sys.exit(1)
-        if parser.has_option('main', 'ninstances'):
-            ninstances = parser.getint('main', 'ninstances')
-        if parser.has_option('main', 'njobs'):
-            njobs = parser.getint('main', 'njobs')
-        if parser.has_option('main', 'http-login'):
-            http_login = parser.get('main', 'http-login')
-        if parser.has_option('main', 'http-password'):
-            http_password = parser.get('main', 'http-password')
-        if parser.has_option('main', 'submitter'):
-            submitter = parser.get('main', 'submitter')
-
-    if args.njobs:
-        njobs = int(args.njobs)
-    if args.ninstances:
-        ninstances = int(args.ninstances)
-    if args.submitter:
-        submitter = args.submitter
-
-    return (ninstances, njobs, http_login, http_password, submitter)
+
+def merge(dict_1, dict_2):
+    """Merge two dictionaries.
+
+    Values that evaluate to true take priority over falsy values.
+    `dict_1` takes priority over `dict_2`.
+
+    """
+    return dict((str(key), dict_1.get(key) or dict_2.get(key))
+                for key in set(dict_2) | set(dict_1))
 
 def main():
     check_version()
     sysinfo = SystemInfo()
-    (ninstances, njobs, http_login, http_password, submitter) = config_get()
-    check_requirements(http_login, http_password)
-    if http_login is None or http_password is None:
+
+    args = docopt.docopt(__doc__, version=VERSION)
+
+    if args['--config']:
+        ini_config = load_ini_config(args['--config'])
+        # merge config/args, priority given to config
+        args = merge(ini_config, args)
+
+    check_requirements(args['--http-login'], args['--http-password'])
+    if args['--http-login'] is None or args['--http-password'] is None:
         print "WARN: due to the lack of http login/password details, results will not be submitted"
         print "WARN: tarballs of results will be kept locally only"
     def sigterm_handler(signum, frame):
         os.killpg(os.getpgid(os.getpid()), signal.SIGTERM)
         sys.exit(1)
     processes = []
-    for i in range(0, ninstances):
-        p = Process(target=run_instance, args=(i, njobs, http_login, http_password, submitter, sysinfo))
+    for i in range(0, int(args['--ninstances'])):
+        p = Process(target=run_instance, args=(i, int(args['--njobs']),
+                args['--http-login'], args['--http-password'],
+                args['--submitter'], sysinfo))
         p.start()
         processes.append(p)
     signal.signal(signal.SIGTERM, sigterm_handler)
Signature · 1 line
1.8.5.1
#3 lore raw

PATCH v5 2/11 autobuild-run: use docopt for argument parsing

Dear Thomas De Schampheleire,

Quoted 9 lines · Thomas De Schampheleire

From: Thomas De Schampheleire <[email protected]>

Using docopt, argument parsing becomes trivial. Adding a new argument is a matter of updating the usage text.
This commit removes the original cumbersome argument handling and uses docopt instead. A method is added to import the settings from the configuration file in a similar dictionary as the one created by docopt, so that both can be merged (giving priority to the configuration file, as before).

Actually, in the current implementation, priority is given to the command line option, and not the configuration file. Like most Unix tools, I believe.

So to make your implementation behave like the existing one, I believe I just need to change:

scripts/autobuild-run def run_instance(instance, njobs, http_login, http_password, submitter, sysinfo) · line 596

Quoted 4 lines
scripts/autobuild-run +4 -0
+    if args['--config']:
+        ini_config = load_ini_config(args['--config'])
+        # merge config/args, priority given to config
+        args = merge(ini_config, args)

to:

	   # merge config/args, priority given to args
	   args = merge(args, ini_config)

No need to resend just for this, I can fixup when applying.

Thanks!

Thomas

Signature · 3 lines
Thomas Petazzoni, CTO, Free Electrons
Embedded Linux, Kernel and Android engineering
http://free-electrons.com
#4 lore raw

PATCH v5 2/11 autobuild-run: use docopt for argument parsing

Quoted 25 lines · Thomas Petazzoni

Dear Thomas De Schampheleire,

Quoted 9 lines · Thomas De Schampheleire

From: Thomas De Schampheleire <[email protected]>

Using docopt, argument parsing becomes trivial. Adding a new argument is a matter of updating the usage text.
This commit removes the original cumbersome argument handling and uses docopt instead. A method is added to import the settings from the configuration file in a similar dictionary as the one created by docopt, so that both can be merged (giving priority to the configuration file, as before).

Actually, in the current implementation, priority is given to the command line option, and not the configuration file. Like most Unix tools, I believe.

So to make your implementation behave like the existing one, I believe I just need to change:

scripts/autobuild-run def run_instance(instance, njobs, http_login, http_password, submitter, sysinfo) · line 596

Quoted 4 lines
scripts/autobuild-run +4 -0
+    if args['--config']:
+        ini_config = load_ini_config(args['--config'])
+        # merge config/args, priority given to config
+        args = merge(ini_config, args)
Quoted 5 lines

to:

           # merge config/args, priority given to args
           args = merge(args, ini_config)

Yes, that should be it.
I had indeed misinterpreted the original code priority.

Thanks,
Thomas

#5 lore raw

PATCH v5 2/11 autobuild-run: use docopt for argument parsing

In reply to Thomas De Schampheleire

Dear Thomas De Schampheleire,

scripts/autobuild-run def run_instance(instance, njobs, http_login, http_password, submitter, sysinfo) · line 596

Quoted 4 lines · Thomas De Schampheleire
scripts/autobuild-run +4 -0
+    if args['--config']:
+        ini_config = load_ini_config(args['--config'])
+        # merge config/args, priority given to config
+        args = merge(ini_config, args)

As we discussed, I changed this to:

args = merge(args, ini_config)

since the command line arguments should have priority over the values defined in the configuration file, like any other Unix tool does.

However, this doesn't seem to work: since the docopt documentation says that ninstances default value is 1 when no value is defined, I believe 'args' contains a value of 1 for ninstances, and it doesn't fall back to using the value defined in the configuration file.

Do you see a way of fixing this?

Thanks,

Thomas

Signature · 3 lines
Thomas Petazzoni, CTO, Free Electrons
Embedded Linux, Kernel and Android engineering
http://free-electrons.com
#6 lore raw

PATCH v5 3/11 autobuild-run: add option --make-opts for custom Buildroot options

In reply to Thomas De Schampheleire

From: Thomas De Schampheleire <[email protected]>

In some environments, the Buildroot configuration may need minor tweaking. For example, in corporate environments behind a firewall, version control commands like git, svn, ... may actually need wrappers. Such wrappers could be configured using the corresponding BR2_<CMD> configuration options.

Since these tweaks are very environment-specific, they should not be part of the configuration file itself. Otherwise, other people trying to reproduce the build in another environment will see builds failing due to other reasons, for example the wrapper not being found.

This commit provides a new configuration option 'make-opts', passable as argument or in the config file, that accepts a string of make options to pass to the Buildroot make command. For example:

    --make-opts="BR2_GIT=git-wrapper BR2_svn=svn-wrapper"
---
 scripts/autobuild-run | 18 ++++++++++++------
 1 file changed, 12 insertions(+), 6 deletions(-)
scripts/autobuild-run +12 -6
index b686a78..8b0091d 100755
@@ -72,6 +72,9 @@ Options:
                                  reasons it is recommended to define this in the
                                  config file instead, with user-read permissions
                                  only)
+  --make-opts OPTSTRING          string of extra options to pass to Buildroot
+                                 make, such as specific command wrappers
+                                 [default: ]
   -c, --config CONFIG            path to configuration file
 
 Format of the configuration file:
@@ -420,7 +423,7 @@ def gen_config(instance, log, sysinfo):
 
     return 0
 
-def do_build(instance, njobs, log):
+def do_build(instance, njobs, log, make_opts):
     """Run the build itself"""
 
     idir = "instance-%d" % instance
@@ -433,8 +436,10 @@ def do_build(instance, njobs, log):
     srcdir = os.path.join(idir, "buildroot")
     f = open(os.path.join(outputdir, "logfile"), "w+")
     log_write(log, "INFO: build started")
-    ret = subprocess.call(["timeout", str(MAX_DURATION), "make", "O=%s" % outputdir, "-C", srcdir,
-                           "BR2_DL_DIR=%s" % dldir, "BR2_JLEVEL=%s" % njobs], stdout=f, stderr=f)
+    cmd = ["timeout", str(MAX_DURATION), "make", "O=%s" % outputdir,
+            "-C", srcdir, "BR2_DL_DIR=%s" % dldir, "BR2_JLEVEL=%s" % njobs] \
+          + make_opts.split()
+    ret = subprocess.call(cmd, stdout=f, stderr=f)
     # 124 is a special error code that indicates we have reached the
     # timeout
     if ret == 124:
@@ -525,7 +530,8 @@ def send_results(instance, http_login, http_password, submitter, log, result):
         os.rename(os.path.join(outputdir, "results.tar.bz2"), resultfilename)
         log_write(log, "INFO: results saved as %s" % resultfilename)
 
-def run_instance(instance, njobs, http_login, http_password, submitter, sysinfo):
+def run_instance(instance, njobs, http_login, http_password, submitter,
+                 make_opts, sysinfo):
     """Main per-instance loop
 
     Prepare the build, generate a configuration, run the build, and submit the
@@ -551,7 +557,7 @@ def run_instance(instance, njobs, http_login, http_password, submitter, sysinfo)
         if ret != 0:
             continue
 
-        ret = do_build(instance, njobs, instance_log)
+        ret = do_build(instance, njobs, instance_log, make_opts)
         send_results(instance, http_login, http_password, submitter, instance_log, ret)
 
 # args / config file merging inspired by:
@@ -606,7 +612,7 @@ def main():
     for i in range(0, int(args['--ninstances'])):
         p = Process(target=run_instance, args=(i, int(args['--njobs']),
                 args['--http-login'], args['--http-password'],
-                args['--submitter'], sysinfo))
+                args['--submitter'], args['--make-opts'], sysinfo))
         p.start()
         processes.append(p)
     signal.signal(signal.SIGTERM, sigterm_handler)
Signature · 1 line
1.8.5.1
#7 lore raw

PATCH v5 3/11 autobuild-run: add option --make-opts for custom Buildroot options

Dear Thomas De Schampheleire,

scripts/autobuild-run def do_build(instance, njobs, log): · line 441

Quoted 3 lines · Thomas De Schampheleire
scripts/autobuild-run +3 -0
+    cmd = ["timeout", str(MAX_DURATION), "make", "O=%s" % outputdir,
+            "-C", srcdir, "BR2_DL_DIR=%s" % dldir, "BR2_JLEVEL=%s" % njobs] \
+          + make_opts.split()

This breaks badly when no make_opts are passed:

Traceback (most recent call last):
File "/usr/lib/python2.6/multiprocessing/process.py", line 232, in _bootstrap

    self.run()
  File "/usr/lib/python2.6/multiprocessing/process.py", line 88, in run
    self._target(*self._args, **self._kwargs)

File "./autobuild-run", line 660, in run_instance
ret = do_build(**kwargs)
File "./autobuild-run", line 471, in do_build
+ kwargs['make_opts'].split()
AttributeError: 'NoneType' object has no attribute 'split'

Best regards,

Thomas

Signature · 3 lines
Thomas Petazzoni, CTO, Free Electrons
Embedded Linux, Kernel and Android engineering
http://free-electrons.com
#8 lore raw

PATCH v5 3/11 autobuild-run: add option --make-opts for custom Buildroot options

Quoted 12 lines · Thomas Petazzoni

Dear Thomas De Schampheleire,

scripts/autobuild-run def do_build(instance, njobs, log): · line 441

Quoted 3 lines · Thomas De Schampheleire
scripts/autobuild-run +3 -0
+    cmd = ["timeout", str(MAX_DURATION), "make", "O=%s" % outputdir,
+            "-C", srcdir, "BR2_DL_DIR=%s" % dldir, "BR2_JLEVEL=%s" % njobs] \
+          + make_opts.split()

This breaks badly when no make_opts are passed:

Traceback (most recent call last):
File "/usr/lib/python2.6/multiprocessing/process.py", line 232, in _bootstrap

Quoted 9 lines
    self.run()
  File "/usr/lib/python2.6/multiprocessing/process.py", line 88, in run
    self._target(*self._args, **self._kwargs)

File "./autobuild-run", line 660, in run_instance
ret = do_build(**kwargs)
File "./autobuild-run", line 471, in do_build
+ kwargs['make_opts'].split()
AttributeError: 'NoneType' object has no attribute 'split'

make-opts is given the empty string '' as default in the original code I sent, and thus can be splitted without the above error. I guess you are seeing this after making some other change?

#9 lore raw

PATCH v5 3/11 autobuild-run: add option --make-opts for custom Buildroot options

Dear Thomas De Schampheleire,

Quoted 3 lines · Thomas De Schampheleire

make-opts is given the empty string '' as default in the original code I sent, and thus can be splitted without the above error. I guess you are seeing this after making some other change?

Nope, I don't think I've done any change regarding this, unless I did a mistake, of course. You can see the code as I tested it in the Git repository I pointed at.

Thomas

Signature · 3 lines
Thomas Petazzoni, CTO, Free Electrons
Embedded Linux, Kernel and Android engineering
http://free-electrons.com
#10 lore raw

PATCH v5 4/11 autobuild-run: use **kwargs to avoid explicit parameter passthroughs

In reply to Thomas De Schampheleire

From: Thomas De Schampheleire <[email protected]>

The current version of autobuild-run has some extensive explicit parameter passing, for example:

- fixup_config needs sysinfo, passed via gen_config, in turn via run_instance, in turn via main.
- send_results needs username/password settings, passed via run_instance, in turn via main.

Everytime a leaf function needs an extra parameter (for example coming from the arguments or config file), the entire call chain needs to be adapted to pass along that parameter.

This patch introduces the **kwargs dictionary principle, that allows implicit parameter passing. A function can accept this dictionary and extract parameters from it by name. The dictionary can be passed as a whole to a child function, without explicitly enumerating which entries in the dictionary are needed in the child.

---

v5: use dict(foo = A, bar = B) instead of { 'foo' = A, 'bar' = B } as this
is more readable.

 scripts/autobuild-run | 77 +++++++++++++++++++++++++++++++--------------------
 1 file changed, 47 insertions(+), 30 deletions(-)
scripts/autobuild-run +47 -30
index 8b0091d..4fc883a 100755
@@ -189,7 +189,7 @@ def get_toolchain_configs():
         configs.append(config)
     return configs
 
-def prepare_build(instance, log):
+def prepare_build(**kwargs):
     """Prepare for the next build of the specified instance
 
     This function prepares the build by making sure all the needed
@@ -197,7 +197,8 @@ def prepare_build(instance, log):
     code, and cleaning up remaining stuff from previous builds.
     """
 
-    idir = "instance-%d" % instance
+    idir = "instance-%d" % kwargs['instance']
+    log = kwargs['log']
 
     log_write(log, "INFO: preparing a new build")
 
@@ -246,7 +247,7 @@ def prepare_build(instance, log):
 
     return 0
 
-def fixup_config(instance, sysinfo):
+def fixup_config(**kwargs):
     """Finalize the configuration and reject any problematic combinations
 
     This function returns 'True' when the configuration has been
@@ -255,9 +256,10 @@ def fixup_config(instance, sysinfo):
     generated).
     """
 
-    idir = "instance-%d" % instance
-    outputdir = os.path.join(idir, "output")
+    idir = "instance-%d" % kwargs['instance']
+    sysinfo = kwargs['sysinfo']
 
+    outputdir = os.path.join(idir, "output")
     with open(os.path.join(outputdir, ".config")) as configf:
         configlines = configf.readlines()
 
@@ -347,7 +349,7 @@ def fixup_config(instance, sysinfo):
 
     return True
 
-def gen_config(instance, log, sysinfo):
+def gen_config(**kwargs):
     """Generate a new random configuration
 
     This function generates the configuration, by choosing a random
@@ -355,7 +357,9 @@ def gen_config(instance, log, sysinfo):
     packages.
     """
 
-    idir = "instance-%d" % instance
+    idir = "instance-%d" % kwargs['instance']
+    log = kwargs['log']
+
     dldir = os.path.join(idir, "dl")
     # We need the absolute path to use with O=, because the relative
     # path to the output directory here is not relative to the
@@ -406,7 +410,7 @@ def gen_config(instance, log, sysinfo):
         if ret != 0:
             log_write(log, "ERROR: cannot generate random configuration")
             return -1
-        if fixup_config(instance, sysinfo):
+        if fixup_config(**kwargs):
             break
 
     ret = subprocess.call(["yes '' 2>/dev/null| make O=%s -C %s oldconfig" % \
@@ -423,10 +427,12 @@ def gen_config(instance, log, sysinfo):
 
     return 0
 
-def do_build(instance, njobs, log, make_opts):
+def do_build(**kwargs):
     """Run the build itself"""
 
-    idir = "instance-%d" % instance
+    idir = "instance-%d" % kwargs['instance']
+    log = kwargs['log']
+
     # We need the absolute path to use with O=, because the relative
     # path to the output directory here is not relative to the
     # Buildroot sources, but to the location of the autobuilder
@@ -437,8 +443,9 @@ def do_build(instance, njobs, log, make_opts):
     f = open(os.path.join(outputdir, "logfile"), "w+")
     log_write(log, "INFO: build started")
     cmd = ["timeout", str(MAX_DURATION), "make", "O=%s" % outputdir,
-            "-C", srcdir, "BR2_DL_DIR=%s" % dldir, "BR2_JLEVEL=%s" % njobs] \
-          + make_opts.split()
+            "-C", srcdir, "BR2_DL_DIR=%s" % dldir,
+            "BR2_JLEVEL=%s" % kwargs['njobs']] \
+          + kwargs['make_opts'].split()
     ret = subprocess.call(cmd, stdout=f, stderr=f)
     # 124 is a special error code that indicates we have reached the
     # timeout
@@ -455,7 +462,7 @@ def do_build(instance, njobs, log, make_opts):
     log_write(log, "INFO: build successful")
     return 0
 
-def send_results(instance, http_login, http_password, submitter, log, result):
+def send_results(result, **kwargs):
     """Prepare and store/send tarball with results
 
     This function prepares the tarball with the results, and either
@@ -463,7 +470,9 @@ def send_results(instance, http_login, http_password, submitter, log, result):
     are available) or stores them locally as tarballs.
     """
 
-    idir = "instance-%d" % instance
+    idir = "instance-%d" % kwargs['instance']
+    log = kwargs['log']
+
     outputdir = os.path.abspath(os.path.join(idir, "output"))
     srcdir = os.path.join(idir, "buildroot")
     resultdir = os.path.join(outputdir, "results")
@@ -498,7 +507,7 @@ def send_results(instance, http_login, http_password, submitter, log, result):
     resultf.close()
 
     with open(os.path.join(resultdir, "submitter"), "w+") as submitterf:
-        submitterf.write(submitter)
+        submitterf.write(kwargs['submitter'])
 
     # Yes, shutil.make_archive() would be nice, but it doesn't exist
     # in Python 2.6.
@@ -508,11 +517,12 @@ def send_results(instance, http_login, http_password, submitter, log, result):
         log_write(log, "ERROR: could not make results tarball")
         sys.exit(1)
 
-    if http_login and http_password:
+    if kwargs['http_login'] and kwargs['http_password']:
         # Submit results. Yes, Python has some HTTP libraries, but
         # none of the ones that are part of the standard library can
         # upload a file without writing dozens of lines of code.
-        ret = subprocess.call(["curl", "-u", "%s:%s" % (http_login, http_password),
+        ret = subprocess.call(["curl", "-u",
+                               "%s:%s" % (kwargs['http_login'], kwargs['http_password']),
                                "-H", "Expect:",
                                "-F", "uploadedfile=@%s" % os.path.join(outputdir, "results.tar.bz2"),
                                "-F", "uploadsubmit=1",
@@ -526,39 +536,40 @@ def send_results(instance, http_login, http_password, submitter, log, result):
         # No http login/password, keep tarballs locally
         with open(os.path.join(outputdir, "results.tar.bz2"), 'rb') as f:
             sha1 = hashlib.sha1(f.read()).hexdigest()
-        resultfilename = "instance-%d-%s.tar.bz2" % (instance, sha1)
+        resultfilename = "instance-%d-%s.tar.bz2" % (kwargs['instance'], sha1)
         os.rename(os.path.join(outputdir, "results.tar.bz2"), resultfilename)
         log_write(log, "INFO: results saved as %s" % resultfilename)
 
-def run_instance(instance, njobs, http_login, http_password, submitter,
-                 make_opts, sysinfo):
+def run_instance(**kwargs):
     """Main per-instance loop
 
     Prepare the build, generate a configuration, run the build, and submit the
     results.
     """
 
-    idir = "instance-%d" % instance
+    idir = "instance-%d" % kwargs['instance']
 
     # If it doesn't exist, create the instance directory
     if not os.path.exists(idir):
         os.mkdir(idir)
 
-    instance_log = open(os.path.join(idir, "instance.log"), "a+")
-    log_write(instance_log, "INFO: instance started")
+    kwargs['log'] = open(os.path.join(idir, "instance.log"), "a+")
+    log_write(kwargs['log'], "INFO: instance started")
+
     while True:
         check_version()
 
-        ret = prepare_build(instance, instance_log)
+        ret = prepare_build(**kwargs)
         if ret != 0:
             continue
 
-        ret = gen_config(instance, instance_log, sysinfo)
+        ret = gen_config(**kwargs)
         if ret != 0:
             continue
 
-        ret = do_build(instance, njobs, instance_log, make_opts)
-        send_results(instance, http_login, http_password, submitter, instance_log, ret)
+        ret = do_build(**kwargs)
+
+        send_results(ret, **kwargs)
 
 # args / config file merging inspired by:
 # https://github.com/docopt/docopt/blob/master/examples/config_file_example.py
@@ -610,9 +621,15 @@ def main():
         sys.exit(1)
     processes = []
     for i in range(0, int(args['--ninstances'])):
-        p = Process(target=run_instance, args=(i, int(args['--njobs']),
-                args['--http-login'], args['--http-password'],
-                args['--submitter'], args['--make-opts'], sysinfo))
+        p = Process(target=run_instance, kwargs=dict(
+                instance = i,
+                njobs = args['--njobs'],
+                sysinfo = sysinfo,
+                http_login = args['--http-login'],
+                http_password = args['--http-password'],
+                submitter = args['--submitter'],
+                make_opts = args['--make-opts']
+            ))
         p.start()
         processes.append(p)
     signal.signal(signal.SIGTERM, sigterm_handler)
Signature · 1 line
1.8.5.1
#11 lore raw

PATCH v5 4/11 autobuild-run: use **kwargs to avoid explicit parameter passthroughs

Dear Thomas De Schampheleire,

Quoted 14 lines · Thomas De Schampheleire

From: Thomas De Schampheleire <[email protected]>

The current version of autobuild-run has some extensive explicit parameter passing, for example:

- fixup_config needs sysinfo, passed via gen_config, in turn via run_instance, in turn via main.
- send_results needs username/password settings, passed via run_instance, in turn via main.

Everytime a leaf function needs an extra parameter (for example coming from the arguments or config file), the entire call chain needs to be adapted to pass along that parameter.

Quoted 5 lines

This patch introduces the **kwargs dictionary principle, that allows implicit parameter passing. A function can accept this dictionary and extract parameters from it by name. The dictionary can be passed as a whole to a child function, without explicitly enumerating which entries in the dictionary are needed in the child.

I must say I find this **kwargs thing a bit weird. Yes indeed, it avoids passing several parameters to functions. But on the other hand, it means that you're passing the *entire* context to *all* functions. Which more or less boils down to using global variables. It means that it isn't clear looking at the prototype of a function, which parameters it is taking.

Maybe it's just that I'm not doing enough Python, and I'm thinking too much C, but I find this **kwargs to not be such a great idea. But oh well, since this is Python, I guess I should convince myself to use the Python spirit :)

Best regards,

Thomas

Signature · 3 lines
Thomas Petazzoni, CTO, Free Electrons
Embedded Linux, Kernel and Android engineering
http://free-electrons.com
#12 lore raw

PATCH v5 4/11 autobuild-run: use **kwargs to avoid explicit parameter passthroughs

Hi Thomas,

Quoted 30 lines · Thomas Petazzoni

Dear Thomas De Schampheleire,

Quoted 14 lines · Thomas De Schampheleire

From: Thomas De Schampheleire <[email protected]>

The current version of autobuild-run has some extensive explicit parameter passing, for example:

- fixup_config needs sysinfo, passed via gen_config, in turn via run_instance, in turn via main.
- send_results needs username/password settings, passed via run_instance, in turn via main.

Everytime a leaf function needs an extra parameter (for example coming from the arguments or config file), the entire call chain needs to be adapted to pass along that parameter.

Quoted 5 lines

This patch introduces the **kwargs dictionary principle, that allows implicit parameter passing. A function can accept this dictionary and extract parameters from it by name. The dictionary can be passed as a whole to a child function, without explicitly enumerating which entries in the dictionary are needed in the child.

I must say I find this **kwargs thing a bit weird. Yes indeed, it avoids passing several parameters to functions. But on the other hand, it means that you're passing the *entire* context to *all* functions. Which more or less boils down to using global variables. It means that it isn't clear looking at the prototype of a function, which parameters it is taking.

Quoted 5 lines

Maybe it's just that I'm not doing enough Python, and I'm thinking too much C, but I find this **kwargs to not be such a great idea. But oh well, since this is Python, I guess I should convince myself to use the Python spirit :)

Maybe someone else with good Python knowledge can chime in and tell us if this is a good way or not of handling things.
Even in C, a function that takes 4+ arguments is not a good idea. Passing such things in a struct (or a dict in python) is more appropriate, IMO.

Best regards,
Thomas

#13 lore raw

PATCH v5 5/11 autobuild-run: check-requirements does not need to know the login details

In reply to Thomas De Schampheleire

From: Thomas De Schampheleire <[email protected]>

check-requirements simply has to know if the results have to be sent, so it can check on some extra requirements. The username and password are irrelevant here.
This commit introduces a boolean variable upload to hide these details.

---
 scripts/autobuild-run | 18 ++++++++++++------
 1 file changed, 12 insertions(+), 6 deletions(-)
scripts/autobuild-run +12 -6
index 4fc883a..ab53d3f 100755
@@ -120,12 +120,12 @@ def check_version():
         print "ERROR: script version too old, please upgrade."
         sys.exit(1)
 
-def check_requirements(http_login, http_password):
+def check_requirements(upload=False):
     devnull = open(os.devnull, "w")
     needed_progs = ["make", "git", "gcc", "timeout"]
     missing_requirements = False
 
-    if http_login and http_password:
+    if upload:
         needed_progs.append("curl")
 
     for prog in needed_progs:
@@ -517,7 +517,7 @@ def send_results(result, **kwargs):
         log_write(log, "ERROR: could not make results tarball")
         sys.exit(1)
 
-    if kwargs['http_login'] and kwargs['http_password']:
+    if kwargs['upload']:
         # Submit results. Yes, Python has some HTTP libraries, but
         # none of the ones that are part of the standard library can
         # upload a file without writing dozens of lines of code.
@@ -612,10 +612,15 @@ def main():
         # merge config/args, priority given to config
         args = merge(ini_config, args)
 
-    check_requirements(args['--http-login'], args['--http-password'])
-    if args['--http-login'] is None or args['--http-password'] is None:
+    # http_login/password could theoretically be allowed as empty, so check
+    # explicitly on None.
+    upload = (args['--http-login'] is not None) \
+                and (args['--http-password'] is not None)
+    check_requirements(upload)
+    if not upload:
         print "WARN: due to the lack of http login/password details, results will not be submitted"
         print "WARN: tarballs of results will be kept locally only"
+
     def sigterm_handler(signum, frame):
         os.killpg(os.getpgid(os.getpid()), signal.SIGTERM)
         sys.exit(1)
@@ -628,7 +633,8 @@ def main():
                 http_login = args['--http-login'],
                 http_password = args['--http-password'],
                 submitter = args['--submitter'],
-                make_opts = args['--make-opts']
+                make_opts = args['--make-opts'],
+                upload = upload
             ))
         p.start()
         processes.append(p)
Signature · 1 line
1.8.5.1
#14 lore raw

PATCH v5 6/11 autobuild-run: set LC_ALL=C to not use locale settings of host machine

In reply to Thomas De Schampheleire

From: Thomas De Schampheleire <[email protected]>

If the host machine running autobuild-run happens to have set a non-English locale, the error messages will be displayed in that language too. For public results like those generated in the Buildroot autobuilders, this is a problem.

Therefore, set LC_ALL to a fixed locale (C) at the beginning of the script.

---
 scripts/autobuild-run | 7 +++++--
 1 file changed, 5 insertions(+), 2 deletions(-)
scripts/autobuild-run +5 -2
index ab53d3f..d085029 100755
@@ -48,8 +48,6 @@
 #   of just using the last 500 lines of the build log, search the
 #   start of the build of the failing package.
 #
-# - Add LC_ALL=C where appropriate.
-#
 # - Include the config.log file (when it exists) in the tarball for
 #   failed builds when the failure occurs on an autotools package.
 #
@@ -602,6 +600,11 @@ def merge(dict_1, dict_2):
                 for key in set(dict_2) | set(dict_1))
 
 def main():
+
+    # Avoid locale settings of autobuilder machine leaking in, for example
+    # showing error messages in another language.
+    os.environ['LC_ALL'] = 'C'
+
     check_version()
     sysinfo = SystemInfo()
 
Signature · 1 line
1.8.5.1
#15 lore raw

PATCH v5 7/11 autobuild-run: improve the logic to generate build-end.log

In reply to Thomas De Schampheleire

From: Thomas De Schampheleire <[email protected]>

This patch makes the generation of build-end.log more intelligent, by not simply adding the 500 last lines, but instead capture the entire output of the failed package.

If no failure reason can be found (possibly a good run) or the start for that package cannot be found, use the 500 last lines as fallback.

The logic to find the failed package is taken from the result import script web/import.inc.php. If needed, the search range of the last 3 lines could be extended.

Note that the package-version string extracted from the build log is split on the last hyphen into a tuple (package, version). This is needed to find back the beginning of that package's build (where the package name is separated from the version with a space, not a hyphen). However, this logic will fail to work when the version also contains a hyphen, for example lmbench-3.0-a9. In this case, the resulting tuple will be (lmbench-3.0,a9) and the text searched in the build log '>>> lmbench-3.0 a9' (which will not be found). In this case, the fallback of the last 500 lines will be used.

---
 scripts/autobuild-run | 53 ++++++++++++++++++++++++++++++++++++++++++++-------
 1 file changed, 46 insertions(+), 7 deletions(-)
scripts/autobuild-run +46 -7
index d085029..32466e8 100755
@@ -44,10 +44,6 @@
 #
 # TODO:
 #
-# - Improve the logic that generates the 'build-end.log' file. Instead
-#   of just using the last 500 lines of the build log, search the
-#   start of the build of the failing package.
-#
 # - Include the config.log file (when it exists) in the tarball for
 #   failed builds when the failure occurs on an autotools package.
 #
@@ -491,9 +487,52 @@ def send_results(result, **kwargs):
     subprocess.call(["git log master -n 1 --pretty=format:%%H > %s" % \
                      os.path.join(resultdir, "gitid")],
                     shell=True, cwd=srcdir)
-    subprocess.call(["tail -500 %s > %s" % \
-                     (os.path.join(outputdir, "logfile"), os.path.join(resultdir, "build-end.log"))],
-                    shell=True)
+
+    def get_failure_reason():
+        # Output is a tuple (package, version), or None.
+
+        lastlines = subprocess.check_output(["tail", "-n", "3",
+                        os.path.join(outputdir, "logfile")]).splitlines()
+
+        import re
+        regexp = re.compile(r'make: \*\*\* .*/(?:build|toolchain)/([^/]*)/')
+        for line in lastlines:
+            m = regexp.search(line)
+            if m:
+                return m.group(1).rsplit('-', 1)
+
+        # not found
+        return None
+
+    def extract_end_log(resultfile):
+        """Save the last part of the build log, starting from the failed package"""
+
+        def extract_last_500_lines():
+            subprocess.call(["tail -500 %s > %s" % \
+                             (os.path.join(outputdir, "logfile"), resultfile)],
+                            shell=True)
+
+        reason = get_failure_reason()
+        if not reason:
+            extract_last_500_lines()
+        else:
+            import mmap
+            f = open(os.path.join(outputdir, "logfile"), 'r')
+            mf = mmap.mmap(f.fileno(), 0, access=mmap.ACCESS_READ)
+            mf.seek(0)
+            # Search for first action on the failed package
+            offset = mf.find('>>> %s' % ' '.join(reason))
+            if offset != -1:
+                with open(resultfile, "w") as endlog:
+                    endlog.write(mf[offset:])
+            else:
+                # not found, use last 500 lines as fallback
+                extract_last_500_lines()
+
+            mf.close()
+            f.close()
+
+    extract_end_log(os.path.join(resultdir, "build-end.log"))
 
     resultf = open(os.path.join(resultdir, "status"), "w+")
     if result == 0:
Signature · 1 line
1.8.5.1
#16 lore raw

PATCH v5 8/11 autobuild-run: save config.log files for failed package

In reply to Thomas De Schampheleire

From: Thomas De Schampheleire <[email protected]>

This patch will save all config.log files of the failed package in the result tarball, to help problem analysis. This is typically useful for autotools packages, although the added logic does not check that explicitly (i.e. any failing package that has a config.log file in the output directory will see this file saved).

The saving of config.log files happens recursively, the directory structure is preserved in the result directory.

---
 scripts/autobuild-run | 23 ++++++++++++++++++++---
 1 file changed, 20 insertions(+), 3 deletions(-)
scripts/autobuild-run +20 -3
index 32466e8..198c7bc 100755
@@ -44,9 +44,6 @@
 #
 # TODO:
 #
-# - Include the config.log file (when it exists) in the tarball for
-#   failed builds when the failure occurs on an autotools package.
-#
 # - Instead of excluding all configurations that have
 #   BR2_PACKAGE_CLASSPATH=y, improve the script to detect whether the
 #   necessary host machine requirements are there to build classpath.
@@ -534,6 +531,26 @@ def send_results(result, **kwargs):
 
     extract_end_log(os.path.join(resultdir, "build-end.log"))
 
+    def copy_config_log_files():
+        """Recursively copy any config.log files from the failing package"""
+
+        reason = get_failure_reason()
+        if not reason:
+            return
+
+        srcroot = os.path.join(outputdir, "build", '-'.join(reason))
+        destroot = os.path.join(resultdir, '-'.join(reason))
+
+        for root, dirs, files in os.walk(srcroot):
+            dest = os.path.join(destroot, os.path.relpath(root, srcroot))
+
+            for file in files:
+                if file == 'config.log':
+                    os.makedirs(dest)
+                    shutil.copy(os.path.join(root, file), os.path.join(dest, file))
+
+    copy_config_log_files()
+
     resultf = open(os.path.join(resultdir, "status"), "w+")
     if result == 0:
         resultf.write("OK")
Signature · 1 line
1.8.5.1
#17 lore raw

PATCH v5 9/11 autobuild-run: extend TODO list

In reply to Thomas De Schampheleire

From: Thomas De Schampheleire <[email protected]>

Some misc. additions to the autobuild-run TODO list.

---
 scripts/autobuild-run | 11 +++++++++++
 1 file changed, 11 insertions(+)
scripts/autobuild-run +11 -0
index 198c7bc..237d443 100755
@@ -47,6 +47,17 @@
 # - Instead of excluding all configurations that have
 #   BR2_PACKAGE_CLASSPATH=y, improve the script to detect whether the
 #   necessary host machine requirements are there to build classpath.
+# - Integrate method check-requirements with the SysInfo class, distinghuishing
+#   between required and optional dependencies.
+# - Extend required dependencies to subversion, mercurial, cpio, wget, python,
+#   etc.
+# - Detect selection of multiple virtual package providers and don't consider it
+#   a failure
+# - Properly handle kill of main script: stop all running children and do not
+#   print out stacktraces
+# - Fix problem in removal of output directory: sometimes this fails with
+#   message 'directory not empty' which suggests that someone is writing to the
+#   directory at the time of removal.
 
 """autobuild-run - run Buildroot autobuilder
 
Signature · 1 line
1.8.5.1
#18 lore raw

PATCH v5 10/11 autobuild-run: kill all children on SIGTERM

In reply to Thomas De Schampheleire

From: Thomas De Schampheleire <[email protected]>

The autobuild-run spawns the main build process through the timeout command. To handle its job correctly, this command creates all children in its own process group, different from the process group of autobuild-run itself.
Thus, when autobuild-run is killed and the signal handler kills the entire process group, the build processes run through timeout remain alive.

To handle this, record the PIDs of the timeout processes in an array shared between the main autobuild-run process and its instances. The signal handler will iterate over all active processes in this array, and kill them explicitly.

If a new timeout process would be started after the signal handler was invoked but before the entire process tree is killed, this process could remain alive too. To prevent this from occurring, the signal handler now starts with terminating all instances.

Lastly, the signal handler would be called for all instances, which is not intended, so prevent that by uninstalling the signal handler as a first step of the handler itself.

---
 scripts/autobuild-run | 39 ++++++++++++++++++++++++++++++++++++---
 1 file changed, 36 insertions(+), 3 deletions(-)
scripts/autobuild-run +36 -3
index 237d443..3c448bd 100755
@@ -97,9 +97,10 @@ import urllib2
 import csv
 from random import randint
 import subprocess
-from multiprocessing import Process
+import multiprocessing
 import signal
 import os
+import errno
 import shutil
 from time import localtime, strftime
 import sys
@@ -444,11 +445,16 @@ def do_build(**kwargs):
     srcdir = os.path.join(idir, "buildroot")
     f = open(os.path.join(outputdir, "logfile"), "w+")
     log_write(log, "INFO: build started")
+
     cmd = ["timeout", str(MAX_DURATION), "make", "O=%s" % outputdir,
             "-C", srcdir, "BR2_DL_DIR=%s" % dldir,
             "BR2_JLEVEL=%s" % kwargs['njobs']] \
           + kwargs['make_opts'].split()
-    ret = subprocess.call(cmd, stdout=f, stderr=f)
+    sub = subprocess.Popen(cmd, stdout=f, stderr=f)
+    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:
@@ -692,8 +698,32 @@ def main():
         print "WARN: tarballs of results will be kept locally only"
 
     def sigterm_handler(signum, frame):
+        """Kill all children"""
+
+        # uninstall signal handler to prevent being called for all subprocesses
+        signal.signal(signal.SIGTERM, signal.SIG_DFL)
+
+        # stop all instances to prevent new children to be spawned
+        for p in processes:
+            p.terminate()
+
+        # kill build processes started with timeout (that puts its children
+        # explicitly in a separate process group)
+        for pid in buildpid:
+            if pid == 0:
+                continue
+            try:
+                os.kill(pid, signal.SIGTERM)
+            except OSError as e:
+                if e.errno != errno.ESRCH: # No such process, ignore
+                    raise
+
+        # kill any remaining children in our process group
         os.killpg(os.getpgid(os.getpid()), signal.SIGTERM)
+
         sys.exit(1)
+
+    buildpid = multiprocessing.Array('i', int(args['--ninstances']))
     processes = []
     for i in range(0, int(args['--ninstances'])):
         p = Process(target=run_instance, kwargs=dict(
@@ -704,11 +734,14 @@ def main():
                 http_password = args['--http-password'],
                 submitter = args['--submitter'],
                 make_opts = args['--make-opts'],
-                upload = upload
+                upload = upload,
+                buildpid = buildpid
             ))
         p.start()
         processes.append(p)
+
     signal.signal(signal.SIGTERM, sigterm_handler)
+
     for p in processes:
         p.join()
 
Signature · 1 line
1.8.5.1
#19 lore raw

PATCH v5 10/11 autobuild-run: kill all children on SIGTERM

Quoted 20 lines · Thomas De Schampheleire

From: Thomas De Schampheleire <[email protected]>

The autobuild-run spawns the main build process through the timeout command. To handle its job correctly, this command creates all children in its own process group, different from the process group of autobuild-run itself.
Thus, when autobuild-run is killed and the signal handler kills the entire process group, the build processes run through timeout remain alive.

To handle this, record the PIDs of the timeout processes in an array shared between the main autobuild-run process and its instances. The signal handler will iterate over all active processes in this array, and kill them explicitly.

If a new timeout process would be started after the signal handler was invoked but before the entire process tree is killed, this process could remain alive too. To prevent this from occurring, the signal handler now starts with terminating all instances.

Quoted 8 lines

Lastly, the signal handler would be called for all instances, which is not intended, so prevent that by uninstalling the signal handler as a first step of the handler itself.

---
 scripts/autobuild-run | 39 ++++++++++++++++++++++++++++++++++++---
 1 file changed, 36 insertions(+), 3 deletions(-)

Some notes here:

- I forgot to remove the TODO entry in this patch. This could be done when patches are merged or in a next iteration.

- Thomas Petazzoni previously proposed to get rid of 'timeout' and handle the timeout in Python logic. I did investigate this path but the end result is more complex. The solution would look like this:

1. instead of calling subprocess.call('timeout make ...') from the
instance directly, one would spawn yet another process just for the subprocess call. Inside this process, one would do subprocess.call('make').

2. the new Process would be joined with a timeout (join(timeout))

3. however, while this join will return when the timeout expires,
the process itself will continue to live. This is not what we want. To solve that problem, we have to save the pid of the subprocess call, share it with the instance that spawned the process using multiprocessing.Value, and upon timeout, let the instance kill the spawned subprocess and all its children. Killing the subprocess is not that hard, but I think one would need to do special things to ensure all the children are killed too. In fact, the timeout command already handles all this: it creates a separate process group so that it can kill the entire process group.

Looking at these steps, step 4 is more or less what is done in the
current patch that still uses the 'timeout' command, while steps 1 and 2 are new and make the solution unnecessarily complex. Hence I opted to keep 'timeout'.

Best regards,
Thomas

#20 lore raw

PATCH v5 10/11 autobuild-run: kill all children on SIGTERM

In reply to Thomas De Schampheleire

[now with the buildroot-subscribed address]

Quoted 15 lines · Thomas De Schampheleire

From: Thomas De Schampheleire <[email protected]>

The autobuild-run spawns the main build process through the timeout command. To handle its job correctly, this command creates all children in its own process group, different from the process group of autobuild-run itself.
Thus, when autobuild-run is killed and the signal handler kills the entire process group, the build processes run through timeout remain alive.

To handle this, record the PIDs of the timeout processes in an array shared between the main autobuild-run process and its instances. The signal handler will iterate over all active processes in this array, and kill them explicitly.

Quoted 8 lines

If a new timeout process would be started after the signal handler was invoked but before the entire process tree is killed, this process could remain alive too. To prevent this from occurring, the signal handler now starts with terminating all instances.

Lastly, the signal handler would be called for all instances, which is not intended, so prevent that by uninstalling the signal handler as a first step of the handler itself.

Some notes here:

- I forgot to remove the TODO entry in this patch. This could be done when patches are merged or in a next iteration.

- Thomas Petazzoni previously proposed to get rid of 'timeout' and handle the timeout in Python logic. I did investigate this path but the end result is more complex. The solution would look like this:

1. instead of calling subprocess.call('timeout make ...') from the
instance directly, one would spawn yet another process just for the subprocess call. Inside this process, one would do subprocess.call('make').

2. the new Process would be joined with a timeout (join(timeout))

3. however, while this join will return when the timeout expires,
the process itself will continue to live. This is not what we want. To solve that problem, we have to save the pid of the subprocess call, share it with the instance that spawned the process using multiprocessing.Value, and upon timeout, let the instance kill the spawned subprocess and all its children. Killing the subprocess is not that hard, but I think one would need to do special things to ensure all the children are killed too. In fact, the timeout command already handles all this: it creates a separate process group so that it can kill the entire process group.

Looking at these steps, step 4 is more or less what is done in the
current patch that still uses the 'timeout' command, while steps 1 and 2 are new and make the solution unnecessarily complex. Hence I opted to keep 'timeout'.

Best regards,
Thomas

#21 lore raw

PATCH v5 10/11 autobuild-run: kill all children on SIGTERM

In reply to Thomas De Schampheleire

Dear Thomas De Schampheleire,

scripts/autobuild-run import urllib2 · line 100

Quoted 2 lines · Thomas De Schampheleire
scripts/autobuild-run +1 -1
-from multiprocessing import Process
+import multiprocessing

This change...

Quoted 2 lines
     for i in range(0, int(args['--ninstances'])):
         p = Process(target=run_instance, kwargs=dict(

Breaks this, it should be changed to multiprocessing.Process.

I've fixed that up locally for now, but it essentially means that your patch series was never executed, as it basically cannot work at all :-/

Best regards,

Thomas

Signature · 3 lines
Thomas Petazzoni, CTO, Free Electrons
Embedded Linux, Kernel and Android engineering
http://free-electrons.com
#22 lore raw

PATCH v5 10/11 autobuild-run: kill all children on SIGTERM

Hi Thomas,

Quoted 9 lines · Thomas Petazzoni

Dear Thomas De Schampheleire,

scripts/autobuild-run import urllib2 · line 100

Quoted 2 lines · Thomas De Schampheleire
scripts/autobuild-run +1 -1
-from multiprocessing import Process
+import multiprocessing

This change...

Quoted 7 lines
Quoted 2 lines
     for i in range(0, int(args['--ninstances'])):
         p = Process(target=run_instance, kwargs=dict(

Breaks this, it should be changed to multiprocessing.Process.

I've fixed that up locally for now, but it essentially means that your patch series was never executed, as it basically cannot work at all :-/

Naturally I did test this series, and this patch, but I must have made a final 'fixup' before submitting without retesting. Shame on me, sorry :-s

#23 lore raw

PATCH v5 11/11 autobuild-run: catch KeyboardInterrupt in the same way as SIGTERM

In reply to Thomas De Schampheleire

From: Thomas De Schampheleire <[email protected]>

Do not handle a keyboard interrupt (SIGINT) in any different way than a SIGTERM signal (kill). In both cases, the entire process tree of the autobuild-run process should be killed.

---
 scripts/autobuild-run | 2 ++
 1 file changed, 2 insertions(+)
scripts/autobuild-run +2 -0
index 3c448bd..c8ce918 100755
@@ -701,6 +701,7 @@ def main():
         """Kill all children"""
 
         # uninstall signal handler to prevent being called for all subprocesses
+        signal.signal(signal.SIGINT, signal.SIG_IGN)
         signal.signal(signal.SIGTERM, signal.SIG_DFL)
 
         # stop all instances to prevent new children to be spawned
@@ -740,6 +741,7 @@ def main():
         p.start()
         processes.append(p)
 
+    signal.signal(signal.SIGINT, sigterm_handler)
     signal.signal(signal.SIGTERM, sigterm_handler)
 
     for p in processes:
Signature · 1 line
1.8.5.1
#24 lore raw

v5 cover autobuild-run improvements

In reply to Thomas De Schampheleire

Hi Thomas,

Quoted 53 lines · Thomas De Schampheleire

From: Thomas De Schampheleire <[email protected]>

This series against the buildroot-test repo makes some improvements to the autobuild-run script.

v5:
- add two patches to handle kill/interrupt properly
- make the kwargs dictionary passing to Process more readable
- rebase

v4:
- rename do_send_results into upload (Peter)

v3:
- introduce missing usage of do_send_results (and move patch backwards)
- update TODO list

v2:
- reorder patches
- take into account comments on boolean logic in first patch
- place docopt directly in scripts/ instead of in a subdirectory
- remove patch adding --git (Yann, Thomas)
- some bug fixes in the first version of the patches

As you can see, this iteration still has the docopt patches. Thomas Petazzoni has expressed his reservations against the introduction of docopt, because the benefits wouldn't weigh out against the addition of an external Python module. Thomas told me on IRC he would have a deeper look onto this before making a final decision.

In case the final decision is not to introduce docopt, following improvements could be done to the current code:
- save the arguments from the command-line into a dictionary instead of plain variables
- keep the added ini_config() method to also save the options in a dictionary, and use the added merge() method to merge both arguments and config options into one final configuration, without needing to handle it manually.

Thomas De Schampheleire (11):
scripts: add python module docopt
autobuild-run: use docopt for argument parsing
autobuild-run: add option --make-opts for custom Buildroot options
autobuild-run: use **kwargs to avoid explicit parameter passthroughs
autobuild-run: check-requirements does not need to know the login
details
autobuild-run: set LC_ALL=C to not use locale settings of host machine
autobuild-run: improve the logic to generate build-end.log
autobuild-run: save config.log files for failed package
autobuild-run: extend TODO list
autobuild-run: kill all children on SIGTERM
autobuild-run: catch KeyboardInterrupt in the same way as SIGTERM

Quoted 7 lines
 scripts/autobuild-run      | 350 +++++++++++++++++++--------
 scripts/docopt.LICENSE-MIT |  23 ++
 scripts/docopt.py          | 581 +++++++++++++++++++++++++++++++++++++++++++++
 3 files changed, 851 insertions(+), 103 deletions(-)
 create mode 100644 scripts/docopt.LICENSE-MIT
 create mode 100644 scripts/docopt.py

How about this patchset?

Thanks,
Thomas

#25 lore raw

v5 cover autobuild-run improvements

In reply to Thomas De Schampheleire

Dear Thomas De Schampheleire,

Quoted 2 lines · Thomas De Schampheleire

This series against the buildroot-test repo makes some improvements to the autobuild-run script.

I've finally taken some time to test this series, and I've reported two separate problems as replies to e-mails: the make-opts split() issue, and the merging of argument vs. configuration file values.

To avoid duplicate work, I've pushed the current state of your work (with the fix to the merging order, and the multiprocessing.Process fix) to:

git://git.free-electrons.com/users/thomas-petazzoni/buildroot-test.git

in a branch called 'thomas-ds'.

Since both issues are more or less tied to the docopt migration, can you have a look at them, and maybe simply propose followup patches that I can merge back into the problematic commits, to fix those issues?

Thanks a lot,

Thomas

Signature · 3 lines
Thomas Petazzoni, CTO, Free Electrons
Embedded Linux, Kernel and Android engineering
http://free-electrons.com
#26 lore raw

v5 cover autobuild-run improvements

Hi Thomas,

Quoted 15 lines · Thomas Petazzoni

Dear Thomas De Schampheleire,

Quoted 2 lines · Thomas De Schampheleire

This series against the buildroot-test repo makes some improvements to the autobuild-run script.

I've finally taken some time to test this series, and I've reported two separate problems as replies to e-mails: the make-opts split() issue, and the merging of argument vs. configuration file values.

To avoid duplicate work, I've pushed the current state of your work (with the fix to the merging order, and the multiprocessing.Process fix) to:

Quoted 7 lines

git://git.free-electrons.com/users/thomas-petazzoni/buildroot-test.git

in a branch called 'thomas-ds'.

Since both issues are more or less tied to the docopt migration, can you have a look at them, and maybe simply propose followup patches that I can merge back into the problematic commits, to fix those issues?

Thanks for looking into this.
As replied on both mails, the make-opts split() issue is AFAICS not present in the original code, but the multiprocessing error was a big mistake by me.

I have sent two follow-up patches based on your branch. The first one solves the argument handling priority issue you mentioned, while the second one is a forgotten change in one of the earlier patches, as I had mentioned before.

Let me know if I can address other issues.

Best regards,
Thomas

#27 lore raw

v5 cover autobuild-run improvements

In reply to Thomas De Schampheleire

Dear Thomas De Schampheleire,

Quoted 5 lines · Thomas De Schampheleire

Thomas De Schampheleire (11):
scripts: add python module docopt
autobuild-run: use docopt for argument parsing
autobuild-run: add option --make-opts for custom Buildroot options
autobuild-run: use **kwargs to avoid explicit parameter passthroughs

Quoted 8 lines

autobuild-run: check-requirements does not need to know the login
details
autobuild-run: set LC_ALL=C to not use locale settings of host machine
autobuild-run: improve the logic to generate build-end.log
autobuild-run: save config.log files for failed package
autobuild-run: extend TODO list
autobuild-run: kill all children on SIGTERM
autobuild-run: catch KeyboardInterrupt in the same way as SIGTERM

I've applied your patch series. A few comments, though:

1/ I also applied your two follow-up patches
http://patchwork.ozlabs.org/patch/444865/ and
http://patchwork.ozlabs.org/patch/444866/. As requested in the commit log, the second patch was squashed into the patch adding the support for killing subprocesses.

2/ I had to rework patch http://patchwork.ozlabs.org/patch/444866/
because subprocess.check_output() doesn't exist in Python 2.6. So I've used a different construct.

3/ Killing all subprocesses apparently still doesn't work completely
correctly, I got the backtrace of the attached file once when interrupting the autobuild-run script with Ctrl+C, on a system that uses Python 2.6. And the child processes (timeout processes) were still running.

Thanks a lot, and sorry for the huge delay!

Thomas

Signature · 7 lines
Thomas Petazzoni, CTO, Free Electrons
Embedded Linux, Kernel and Android engineering
http://free-electrons.com
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: backtrace.txt
URL: <http://lists.busybox.net/pipermail/buildroot/attachments/20150315/4339d329/attachment.txt>
#28 lore raw

v5 cover autobuild-run improvements

Hi Thomas,

Quoted 30 lines · Thomas Petazzoni

Dear Thomas De Schampheleire,

Quoted 5 lines · Thomas De Schampheleire

Thomas De Schampheleire (11):
scripts: add python module docopt
autobuild-run: use docopt for argument parsing
autobuild-run: add option --make-opts for custom Buildroot options
autobuild-run: use **kwargs to avoid explicit parameter passthroughs

Quoted 8 lines

autobuild-run: check-requirements does not need to know the login
details
autobuild-run: set LC_ALL=C to not use locale settings of host machine
autobuild-run: improve the logic to generate build-end.log
autobuild-run: save config.log files for failed package
autobuild-run: extend TODO list
autobuild-run: kill all children on SIGTERM
autobuild-run: catch KeyboardInterrupt in the same way as SIGTERM

I've applied your patch series. A few comments, though:

1/ I also applied your two follow-up patches
http://patchwork.ozlabs.org/patch/444865/ and
http://patchwork.ozlabs.org/patch/444866/. As requested in the commit log, the second patch was squashed into the patch adding the support for killing subprocesses.

2/ I had to rework patch http://patchwork.ozlabs.org/patch/444866/
because subprocess.check_output() doesn't exist in Python 2.6. So I've used a different construct.

Quoted 5 lines

3/ Killing all subprocesses apparently still doesn't work completely
correctly, I got the backtrace of the attached file once when interrupting the autobuild-run script with Ctrl+C, on a system that uses Python 2.6. And the child processes (timeout processes) were still running.

To understand better, some questions:
- you saw this situation once, with Python 2.6, but have been able to successfully stop autobuild run with Ctrl-C with Python 2.6 on other occasions?

- did you happen to use Python 2.7 with these patches on a certain machine? Did you see issues?

In my testing I have been using Python 2.7.
It could be that there is something different in Python 2.6, not sure. I would need to look into this deeper, unless Andr? who seems to be much more knowledgeable about Python has an idea?

Thanks,
Thomas

#29 lore raw

v5 cover autobuild-run improvements

Dear Thomas De Schampheleire,

Quoted 4 lines · Thomas De Schampheleire

To understand better, some questions:
- you saw this situation once, with Python 2.6, but have been able to successfully stop autobuild run with Ctrl-C with Python 2.6 on other occasions?

I just tried again, and same behavior. Here are the processes that remain:

22897 pts/2    S      0:00 timeout 14400 make O=/ssd1/thomas/autobuild/instance-0/output -C instance-0/buildroot BR2_DL_DIR=/ssd1/thomas/autobuild/instance-0/dl BR2_JLEVEL=4
22898 pts/2    S      0:15  \_ make O=/ssd1/thomas/autobuild/instance-0/output -C instance-0/buildroot BR2_DL_DIR=/ssd1/thomas/autobuild/instance-0/dl BR2_JLEVEL=4
15337 pts/2    S      0:00      \_ /usr/bin/make -j4 -C /ssd1/thomas/autobuild/instance-0/output/build/host-nodejs-0.10.36 PATH=/ssd1/thomas/autobuild/instance-0/output/host/bin:/ssd1/th
[...]
25240 pts/2    S      0:00 timeout 14400 make O=/ssd1/thomas/autobuild/instance-3/output -C instance-3/buildroot BR2_DL_DIR=/ssd1/thomas/autobuild/instance-3/dl BR2_JLEVEL=4
25242 pts/2    S      0:14  \_ make O=/ssd1/thomas/autobuild/instance-3/output -C instance-3/buildroot BR2_DL_DIR=/ssd1/thomas/autobuild/instance-3/dl BR2_JLEVEL=4
25512 pts/2    S      0:00      \_ /usr/bin/make -j4 -C /ssd1/thomas/autobuild/instance-3/output/build/samba4-4.2.0 DESTDIR=/ssd1/thomas/autobuild/instance-3/output/host/usr/i686-buildro
[...]
 2853 pts/2    S      0:00 timeout 14400 make O=/ssd1/thomas/autobuild/instance-2/output -C instance-2/buildroot BR2_DL_DIR=/ssd1/thomas/autobuild/instance-2/dl BR2_JLEVEL=4
 2854 pts/2    S      0:08  \_ make O=/ssd1/thomas/autobuild/instance-2/output -C instance-2/buildroot BR2_DL_DIR=/ssd1/thomas/autobuild/instance-2/dl BR2_JLEVEL=4
25772 pts/2    S      0:00      \_ /bin/bash -c cd /ssd1/thomas/autobuild/instance-2/output/build/host-heimdal-1.5.3/ &&  PATH="/ssd1/thomas/autobuild/instance-2/output/host/bin:/ssd1/th
[...]
 7028 pts/2    S      0:00 timeout 14400 make O=/ssd1/thomas/autobuild/instance-1/output -C instance-1/buildroot BR2_DL_DIR=/ssd1/thomas/autobuild/instance-1/dl BR2_JLEVEL=4
 7029 pts/2    S      0:06  \_ make O=/ssd1/thomas/autobuild/instance-1/output -C instance-1/buildroot BR2_DL_DIR=/ssd1/thomas/autobuild/instance-1/dl BR2_JLEVEL=4
19878 pts/2    S      0:00      \_ /bin/bash -c (cd /ssd1/thomas/autobuild/instance-1/output/build/host-cmake-3.1.3; LDFLAGS="-L/ssd1/thomas/autobuild/instance-1/output/host/lib -L/ssd1/
Quoted 2 lines

- did you happen to use Python 2.7 with these patches on a certain machine? Did you see issues?

No, I have been testing them only on gcc75 for now, which uses Python 2.6.

I'm planning on deploying them on the Free Electrons build server as well, but it also has Python 2.6.

Quoted 4 lines

In my testing I have been using Python 2.7.
It could be that there is something different in Python 2.6, not sure. I would need to look into this deeper, unless Andr? who seems to be much more knowledgeable about Python has an idea?

That would be nice :-)

Thomas

Signature · 3 lines
Thomas Petazzoni, CTO, Free Electrons
Embedded Linux, Kernel and Android engineering
http://free-electrons.com
#30 lore raw

v5 cover autobuild-run improvements

Hi,

2015-03-15 15:10 GMT+01:00 Thomas Petazzoni
<[email protected]>:

Quoted 12 lines

Dear Thomas De Schampheleire,

Quoted 4 lines · Thomas De Schampheleire

To understand better, some questions:
- you saw this situation once, with Python 2.6, but have been able to successfully stop autobuild run with Ctrl-C with Python 2.6 on other occasions?

I just tried again, and same behavior. Here are the processes that remain:

Quoted 9 lines

[...]

Quoted 4 lines

In my testing I have been using Python 2.7.
It could be that there is something different in Python 2.6, not sure. I would need to look into this deeper, unless Andr? who seems to be much more knowledgeable about Python has an idea?

That would be nice :-)

I'll look into that - one thing to consider is that SIGINT of any form gets also delivered to multiprocessing.Process instances, so putting a "signal.signal(signal.SIGINT, signal.SIG_IGN)" before creating the processes might help (line ~803; ugly workaround, actually). But I doubt it's as easy as that..

Signature · 1 line
Andr?