Skip to contentSkip to conversation

autobuild-run improvements

PATCH Superseded Thomas De Schampheleire 3 participants 18 messages Nov 12, 2014 – Dec 5, 2014

v4 of 5 Superseded Latest is v5 →

lore
Share & tools
lore
#0 lore raw

v4 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.

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 (9):
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

 scripts/autobuild-run      | 309 ++++++++++++++++--------
 scripts/docopt.LICENSE-MIT |  23 ++
 scripts/docopt.py          | 581 +++++++++++++++++++++++++++++++++++++++++++++
 3 files changed, 811 insertions(+), 102 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 v4 1/9 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 v4 2/9 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 13752f9..be4a9e2 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 v4 3/9 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 be4a9e2..e0a87fa 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
#4 lore raw

PATCH v4 4/9 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.

---
 scripts/autobuild-run | 73 ++++++++++++++++++++++++++++++---------------------
 1 file changed, 43 insertions(+), 30 deletions(-)
scripts/autobuild-run +43 -30
index e0a87fa..8312bfb 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 +621,11 @@ 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={
+            '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
#5 lore raw

PATCH v4 5/9 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 8312bfb..3f5b4cf 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)
@@ -625,7 +630,8 @@ def main():
             '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']})
+            'submitter': args['--submitter'], 'make_opts': args['--make-opts'],
+            'upload': upload})
         p.start()
         processes.append(p)
     signal.signal(signal.SIGTERM, sigterm_handler)
Signature · 1 line
1.8.5.1
#6 lore raw

PATCH v4 6/9 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 3f5b4cf..1ef3f8d 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
#7 lore raw

PATCH v4 7/9 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 1ef3f8d..a88838f 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
#8 lore raw

PATCH v4 8/9 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 a88838f..7fe0b00 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
#9 lore raw

PATCH v4 9/9 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 7fe0b00..a73ccc9 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
#10 lore raw

PATCH v4 9/9 autobuild-run: extend TODO list

Hi,

Quoted 15 lines · 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 +0 -0
index 7fe0b00..a73ccc9 100755
@@ -47,6 +47,17 @@
 # - Instead of excluding all configurations that have

scripts/autobuild-run line 57

Quoted 10 lines
scripts/autobuild-run +8 -0
 #   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

Brief update on this topic: the autobuild-run code currently kills the process group associated with the python process, but it turns out that the build jobs are not part of the same process group as the python process.
As far as I can see, the subprocess class should spawn processes into the same process group, so maybe make is explicitly changing this? Further investigation is needed here, but I think that if we can make sure that these processes are part of the same group as the python parent process, then the kill of autobuild-run will automatically kill the children, as expected.

To visualize the process groups: ps -e -o pid,user,pgid,args --sort pgid

Best regards,
Thomas

#11 lore raw

PATCH v4 9/9 autobuild-run: extend TODO list

Dear Thomas De Schampheleire,

Quoted 28 lines · Thomas De Schampheleire
Quoted 6 lines
scripts/autobuild-run +0 -0
index 7fe0b00..a73ccc9 100755
@@ -47,6 +47,17 @@
 # - Instead of excluding all configurations that have

scripts/autobuild-run line 57

Quoted 10 lines
scripts/autobuild-run +8 -0
 #   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

Brief update on this topic: the autobuild-run code currently kills the process group associated with the python process, but it turns out that the build jobs are not part of the same process group as the python process.
As far as I can see, the subprocess class should spawn processes into the same process group, so maybe make is explicitly changing this? Further investigation is needed here, but I think that if we can make sure that these processes are part of the same group as the python parent process, then the kill of autobuild-run will automatically kill the children, as expected.

Quoted 1 line

To visualize the process groups: ps -e -o pid,user,pgid,args --sort pgid

Thanks for this investigation. It would indeed be really nice to be able to kill just autobuild-run, and be sure that it stops all underlying builds. We could also catch the signal in autobuild-run, and kill the build jobs, if we are able to record the necessary information about them.

Thomas

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

PATCH v4 9/9 autobuild-run: extend TODO list

Quoted 32 lines · Thomas Petazzoni

Dear Thomas De Schampheleire,

Quoted 28 lines · Thomas De Schampheleire
Quoted 6 lines
scripts/autobuild-run +0 -0
index 7fe0b00..a73ccc9 100755
@@ -47,6 +47,17 @@
 # - Instead of excluding all configurations that have

scripts/autobuild-run line 57

Quoted 10 lines
scripts/autobuild-run +8 -0
 #   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

Brief update on this topic: the autobuild-run code currently kills the process group associated with the python process, but it turns out that the build jobs are not part of the same process group as the python process.
As far as I can see, the subprocess class should spawn processes into the same process group, so maybe make is explicitly changing this? Further investigation is needed here, but I think that if we can make sure that these processes are part of the same group as the python parent process, then the kill of autobuild-run will automatically kill the children, as expected.

Quoted 8 lines
Quoted 1 line

To visualize the process groups: ps -e -o pid,user,pgid,args --sort pgid

Thanks for this investigation. It would indeed be really nice to be able to kill just autobuild-run, and be sure that it stops all underlying builds. We could also catch the signal in autobuild-run, and kill the build jobs, if we are able to record the necessary information about them.

Brief update: the piece of the puzzle that is causing a new process group to be spawned is the 'timeout' command. Everything started under the timeout command runs in its own process group, different than the one of the python process. Therefore, killing the python process group does not kill the timeout+make command.

The solution is to use subprocess.Popen instead of subprocess.call so we can obtain the pid of the timeout call, and change the signal handler to explicitly kill this second process group too.

I still have to make the changes in autobuild-run (currently worked on a simple test program), and will submit patches when ready...

Best regards,
Thomas

#13 lore raw

PATCH v4 9/9 autobuild-run: extend TODO list

Dear Thomas De Schampheleire,

Quoted 12 lines · Thomas De Schampheleire
Quoted 5 lines

Thanks for this investigation. It would indeed be really nice to be able to kill just autobuild-run, and be sure that it stops all underlying builds. We could also catch the signal in autobuild-run, and kill the build jobs, if we are able to record the necessary information about them.

Brief update: the piece of the puzzle that is causing a new process group to be spawned is the 'timeout' command. Everything started under the timeout command runs in its own process group, different than the one of the python process. Therefore, killing the python process group does not kill the timeout+make command.

Quoted 6 lines

The solution is to use subprocess.Popen instead of subprocess.call so we can obtain the pid of the timeout call, and change the signal handler to explicitly kill this second process group too.

I still have to make the changes in autobuild-run (currently worked on a simple test program), and will submit patches when ready...

Ok, great. Another approach maybe would be to not use timeout, and instead make the Python script itself monitor the duration of the build it starts, and kill them if they took too long.

Thanks again for the investigation!

Thomas

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

PATCH v4 9/9 autobuild-run: extend TODO list

Quoted 20 lines · Thomas Petazzoni

Dear Thomas De Schampheleire,

Quoted 16 lines · Thomas De Schampheleire
Quoted 5 lines

Thanks for this investigation. It would indeed be really nice to be able to kill just autobuild-run, and be sure that it stops all underlying builds. We could also catch the signal in autobuild-run, and kill the build jobs, if we are able to record the necessary information about them.

Brief update: the piece of the puzzle that is causing a new process group to be spawned is the 'timeout' command. Everything started under the timeout command runs in its own process group, different than the one of the python process. Therefore, killing the python process group does not kill the timeout+make command.

The solution is to use subprocess.Popen instead of subprocess.call so we can obtain the pid of the timeout call, and change the signal handler to explicitly kill this second process group too.

Quoted 7 lines
Quoted 2 lines

I still have to make the changes in autobuild-run (currently worked on a simple test program), and will submit patches when ready...

Ok, great. Another approach maybe would be to not use timeout, and instead make the Python script itself monitor the duration of the build it starts, and kill them if they took too long.

The multiprocessing.Process class we're already using to spawn instances, supports timeout in the join() function. So one way is to spawn a Process for the make call (which in itself calls subprocess.call, as today), then joining it with the timeout.

I will try that out, thanks for the hint...

#15 lore raw

v4 cover autobuild-run improvements

In reply to Thomas De Schampheleire

Hi Thomas,

Quoted 25 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.

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.

Quoted 7 lines

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.

Have you made a decision regarding the use of docopt in autobuild-run, and these patches in general?

Thanks,
Thomas

#16 lore raw

v4 cover autobuild-run improvements

Hello,

Quoted 10 lines · Thomas De Schampheleire
Quoted 7 lines

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.

Have you made a decision regarding the use of docopt in autobuild-run, and these patches in general?

No, not yet. It's still on my TODO-list though, but it will probably be done post-2014.11. Sorry for the long delay :/

Thomas

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

v4 cover autobuild-run improvements

Quoted 12 lines · Thomas Petazzoni

Hello,

Quoted 8 lines · Thomas De Schampheleire
Quoted 7 lines

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.

Quoted 6 lines
Quoted 2 lines

Have you made a decision regarding the use of docopt in autobuild-run, and these patches in general?

No, not yet. It's still on my TODO-list though, but it will probably be done post-2014.11. Sorry for the long delay :/

No problem, I've been off of buildroot development the past month anyway, should normally ramp up again now...

Best regards,
Thomas