We want to be able to track down any left-behind tmp files,
so ensure we're creating them with according file names.
Change-Id: I1aa0ccdb8834e8aef919f1f028028c1bb14cbd01
* We were missing the scenario when only the build files for an instance
had changes. The service for that instance was not restarted
Change-Id: I85ba86bc57c7fc8b6df1ac2ef1144c2d3952a4da
* expects values from CONFIG_POOL
* append OUTPUT_DIRECTORY just when changing directory
added at 83003649e3
Change-Id: I9c22bff70b58d4cfb49c3dab45330bccb898011e
Switch all code to use a loop to parse options so that these can parse
multiple options (currently some scripts would simply parse one or
another but not several), and to make them easier to extend.
Change-Id: I5e902f2d7d8db78e66e48db1237924fa7d8fdda1
we need to produce different output for each instance defined
from the same source
* helper/instances-info, produces the needed output to feed
helper/tt2-process with the proper arguments to build
instances files. We inject ``instance_name`` with the
proper value to templates and INSTANCE_NAME to hooks
environment
* ${NGCPCTL_MAIN}/ngcpcfg.instances is where we define the
supported instance templates and where the output path is
* helper/tt2-process, add -r --replace <path:path> option
in order to allow generation of different output from
the same source
Change-Id: I067266b5226485ea0d6c3bc3436275c5553a8177
We need to flush the queues unconditionally regardless on whether we
have executed .services files, otherwise the enqueued services during
package upgrades might be left unattended for long periods of time in
case no .services files got triggered due to modified configuration
files.
We flush all queues in case some enqueued service was left behind in the
default queue.
Change-Id: Iea8ab11fcb7e580cb8e160b8674e2dff3c9c85a0
Like we were doing with heartbeat-2 we need to special case
corosync/pacemaker and force their services first thing, so that the
HA system is ready by the time the rest of the services are handled.
Change-Id: I3215aed2172436e61b6b33c4a4159c1c3f89bc24
Reported-by: Alex Lutay <alutay@sipwise.com>
Quoting from "man bash" about `-E` (AKA errtrace):
| If set, any trap on ERR is inherited by shell functions, command
| substitutions, and commands executed in a subshell environment.
| The ERR trap is normally not inherited in such cases.
To demonstrate the problem see this short shell script:
| % cat foo
| set -eu -o pipefail
|
| bailout() {
| echo "Bailing out because of error" >&2
| exit 1
| }
| trap bailout 1 2 3 6 9 14 15 ERR
|
| foo() {
| echo "Executing magic"
| magic
| }
|
| foo
| echo end
If "magic" can't be executed, then this fails as follows:
| % bash ./foo
| Executing magic
| ./foo: line 11: magic: command not found
But it doesn't invoke the bailout function via trap.
When using `set -eE` (AKA errexit + errtrace), instead of only
`set -e` (errexit), then it behaves as expected though:
| % bash ./foo
| Executing magic
| ./foo: line 11: magic: command not found
| Bailing out because of error
Change-Id: I9f05355716551cd8fc8f22bd2fa16bbc9d0b7247
shellcheck v0.7.1 complains about a bunch of issues:
SC1083: This { is literal. Check expression (missing ;/\n?) or quote it.
SC1083: This } is literal. Check expression (missing ;/\n?) or quote it.
SC1090: Can't follow non-constant source. Use a directive to specify location.
SC2034: foo appears unused. Verify it or export it.
SC2059: Don't use variables in the printf format string. Use printf '..%s..' "$foo".
SC2128: Expanding an array without an index only gives the first element.
SC2148: Tips depend on target shell and yours is unknown. Add a shebang or a 'shell' directive.
SC2155: Declare and assign separately to avoid masking return values.
SC2162: read without -r will mangle backslashes.
SC2164: Use 'cd ... || exit' or 'cd ... || return' in case cd fails.
SC2181: Check exit code directly with e.g. 'if mycmd;', not indirectly with $?.
SC2207: Prefer mapfile or read -a to split command output (or quote to avoid splitting).
SC2237: Use [ -z .. ] instead of ! [ -n .. ].
We're still bound to support shellcheck v0.3.4 and sadly this version
fails hard when using the modern "shellcheck source=" directives to
address SC1090 and SC1091. It's failing like:
| not ok 1 source/functions/init:8:28: note: The mentioned parser error was in this then clause. [SC1009]
| not ok 2 source/functions/init:10:3: error: Couldn't parse this shellcheck annotation. [SC1073]
| not ok 3 source/functions/init:10:16: error: Unexpected "s". Fix any mentioned problems and try again. [SC1072]
So until we can get rid of supporting this ancient shellcheck version,
instead of using:
| # shellcheck source=etc/ngcp-config/ngcpcfg.cfg
| # shellcheck source=functions/init
| # shellcheck source=functions/logs
| # shellcheck source=functions/main
we need to place disable=SC1090 and disable=SC1091 accordingly.
Change-Id: I14da87a00102609bffdb60e932045180c6d13d5e
These services files just setup some sysctl settings and should always
be safe to be executed first. In this case we do require them to be
executed first, because many of our services are binding into the shared
IP, which will not be present on the standby node, and would otherwise
fail.
Change-Id: I0df095403461268cdff8517ee5c369355ca9228c
We already know what we have processed and what remains from the
complete list and the new partial list. Use that instead of hard-coding
the directories, which is error-prone as it requires keeping both lists
in sync.
Change-Id: Ia16ac6999beaf89f4c2fb5b74216d0bf17de8414
We will gather the actions from within policy-rc.d and then execute all
of them in a single batch.
This will make sure any required restart or reload is performed even if
there has been no template file changes.
Change-Id: I6458936860a86ddb10cd738a8f2770c109d9ed37
We should print the deferred service actions that will be performed
before executing them. To add traceability.
Change-Id: I912af91a6b609d20a312182b18f4d959a159de65
The previous logic had several issues that need to be addressed, when
doing an early exit:
- The temporary file was not getting removed.
- The services sync-state step was not being executed.
This makes the code more future-proof, and fixes the above problems,
by moving the sync-state into its own function, removing the early
exit, and making the service explicit state changes conditional on
the previous early exit conditional.
Change-Id: Ie420cbb94ea6d16a83c7ae3f546cd3eb38435dd8
There was a fix for the similar issue in Mantis ticket #7577,
see commit fc5af3bbb2 for more details.
Unfortunately one corner case was missing there:
"git status" doesn't show untracked files in untracked folders by default.
It requires "-u" option with value "all":
>> man git status
> ...
> -u[<mode>], --untracked-files[=<mode>]
> Show untracked files.
>
> The mode parameter is used to specify the handling of untracked files.
> It is optional: it defaults to all, and if specified,
> it must be stuck to the option (e.g. -uno, but not -u no).
>
> The possible options are:
> · no - Show no untracked files.
> · normal - Shows untracked files and directories.
> · all - Also shows individual files in untracked directories.
The ngcpcfg need to see complete path and filename to untracked file
to properly detect .services files and execute them (if configs were changed):
> root@web01a:/etc# git status
> On branch master
> nothing to commit, working tree clean
> root@web01a:/etc# git status --porcelain | sed 's/^...//'
> root@web01a:/etc# touch test1
> root@web01a:/etc# git status --porcelain | sed 's/^...//'
> test1
> root@web01a:/etc# mkdir test2
> root@web01a:/etc# touch test2/test2_file
> root@web01a:/etc# git status --porcelain | sed 's/^...//'
> test1
> test2/
> root@web01a:/etc# git status -uall --porcelain | sed 's/^...//'
> test1
> test2/test2_file
> root@web01a:/etc#
Change-Id: I18286fd76c511061159f90e8aa9841a6e98c7199
Although the NGCPCFG_RUNNING environment variable might be useful for
some of the subprocesses, it is definitely not useful for anything
that gets started via systemd. In that case we have to transmit this
information via the filesystem.
In the future we should probably extend this with some proper locking,
but for our current problem with ngcp-eaddres we'll still want to
completely avoid running if such file exists, not just prevent
multiple concurrent executions.
Change-Id: I0546620d8ef4263177c5d04f7d2546345c42fbb6
This will make sure any service that should not be running is stopped,
and any that should is started.
Change-Id: I7ab181b66ed3554c8f93f2d9f46dc08ee087a87d
At the end of installer's 'configuration' section we need
to restart all services were touched during the installation
(otherwise we need to restart the node completely).
The option helps us in this case as etckeeper is
not reliable enough during the installation.
P.S. we pass options from 'apply' to 'services' now,
we should remove 'Unsupported option(s) given: $*' check,
otherwise it is failing as 'apply' requires commit message,
which cause error on 'services' level: 'Unsupported option(s) given'
> 2018-09-13 18:08:33: Error: Unsupported option(s) given: my commit message
> 2018-09-13 18:08:34: Did you mean '--dry-run'?
Change-Id: I5dac44cd87dc1fc90f81e4ac3a14af32c00f825e
We need to have some reproducible order of services restart,
otherwise this might lead to different results, depending on the order.
Also unifyer() looks too heavy here since the files can be easily
unified in a moment of writing them into ${TMPFILE}.
Change-Id: I0a683a7d57cd52b74cba0ee9a285f84f77f3dc20
It was initially introduced in the commit:
> commit 7d422b7878
> Author: Victor Seva <vseva@sipwise.com>
> Date: Fri Aug 22 17:03:39 2014 +0200
>
> MT#8763 check for .git before running the services
>
> On a multi directory config, the services depends on git. Check
> first if the git repository exists and warn about it
The subshell here was necessary to isolate 'cd $dir' call.
In the same time the "proper" bash way is to use pushd/popd.
For the history, 'shellcheck disable' was muting the warning:
> This only exits the subshell caused by the (..) group. [SC2106]
Change-Id: I3546fa9615858dbbc4c0f91d74de003e22837d55
Previously 'systemd daemon-reload' and 'systemctl preset-all'
were executed only if changes detected in configs.
Otherwise the early exit happened if no .service files were triggered:
> DEBUG: No services file(s) reported - nothing to do.
Move the code into the function systed_daemon_reload_preset and
execute it on the top of the script. Also print info message
informing users about reloaded systemd and newly preset units.
It should provide better visibility here.
Change-Id: I992af9fb274ea93a37b812a51cebcd7af5c54133
Otherwise the code in between the functions can be easily missed
which cause issues as described in the ticket and will be fixed
in the following commit.
Change-Id: I57decfcbcd41691d35d085b13881e1e6b5208f6e
This will collect all service actions, synthetize them into their
minimal expression and execute them in a single batch per action.
Change-Id: I950d5db32e0ec6327964faac4ce8f15449f90e90
Remove any stale .wants/ relationships as systemd does not always seem
to be able to cleanup those by itself. This makes sure we always start
from a clean slate.
Change-Id: I1b241d4530c45658b7245cdbca33e34dde5beaf6
Under systemd we need to issue "daemon-reload" after applying
templates, to avoid spurious warnings when adding/changing/removing
systemd unit files.
Change-Id: Ifd474bec25b94ee9cb1c247b1390103a29c7b7b3
A random order of service restarts might not work because
a service might depend on specific monit resource(s).
Therefore restart monit before any other service(s).
Change-Id: If6eae434014f00a76fdd66aac8c82dfd69b86f1b
On a multi directory config, the services depends on git. Check
first if the git repository exists and warn about it
Change-Id: I52e335c85ecf4a330da44b3088725623f3c04531
`git diff-index --name-only HEAD` was used because it was the
only reasonable way to check for modified files back in Git 1.5.
Sadly this command can't detect files that didn't exist inside
the Git repository at all yet (so if a file was generated for the
very first time the according services script wasn't executed
necessarily).
Instead nowadays we can rely on `git status --porcelain`, which
wasn't available in Git 1.5 back then yet. Even in Debian/squeeze
there's Git version 1.7.2.5-3 available, so it's save to depend
on it. Updated Git package dependency accordingly, also switch
from transitional git-core package to the actual git one.
Tested-by: Alexander Lutay <alutay@sipwise.com>
Michael Prokop
ngcp-config
Guillem Jover
Alexander Lutay
Manuel Montecelo