seasharp
/
anvil
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

* Reworked the anvil-join-anvil job parsing to help diagnose occassional faults. Also changed a fatal parse error to one that allows the run to be retried.

Browse Source 
Signed-off-by: Digimer <digimer@alteeve.ca>
main
Digimer 4 years ago
parent 0aac9ead8c
commit 5b4bfa747c
3 changed files with 37 additions and 27 deletions
Whitespace
Show all changes Ignore whitespace when comparing lines Ignore changes in amount of whitespace Ignore changes in whitespace at EOL
Split View
Diff Options
Show Stats Download Patch File Download Diff File
  1. 4
      ocf/alteeve/server
  2. 6
      share/words.xml
  3. 54
      tools/anvil-join-anvil

4
ocf/alteeve/server
Unescape View File

@ -199,13 +199,13 @@ $anvil->Log->entry({source => $THIS_FILE, line => __LINE__, 'print' => 0, level
foreach my $key (sort {$a cmp $b} keys %{$anvil->data->{environment}})
{
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 1, list => {
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => {
"environment::${key}" => $anvil->data->{environment}{$key},
}});
}
foreach my $key (sort {$a cmp $b} keys %{$anvil->data->{switches}})
{
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 1, list => {
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 2, list => {
"switches::${key}" => $anvil->data->{switches}{$key},
}});
}

6
share/words.xml
Unescape View File

@ -419,6 +419,7 @@ The attempt to start the servers appears to have failed. The return code '0' was
<key name="error_0305">Unable to connect to the database, unable to manage a server at this time.</key>
<key name="error_0306">Unable to connect to the database, unable to provision a server at this time.</key>
<key name="error_0307">Failed to perform requested task(s) because the requester is not authenticated.</key>
<key name="error_0308"><![CDATA[[ Error ] - The Job: [#!variable!job-uuid!#] appears to have passed malformed data. The raw data was: [#!variable!raw!#]. Expected 'as_machine=<host_type>,manifest_uuid=<manifest_uuid>,anvil_uuid=<anvil_uuid>'. Either the parse failed, or the data was somehow invalid.]]></key>
<!-- Files templates -->
<!-- NOTE: Translating these files requires an understanding of which lines are translatable -->
@ -806,7 +807,7 @@ Failure! The return code: [#!variable!return_code!#] was received ('0' was expec
<key name="job_0073">This machine will join an #!string!brand_0006!# as a node or DR host. The role and #!string!brand_0006!# will be determined by the associated Install Manifest UUID.</key>
<key name="job_0074">'Join #!string!brand_0002!#' job: [#!variable!job-uuid!#] picked up.</key>
<key name="job_0075">This will become: [#!variable!machine!#] using data from the install manifest UUID: [#!variable!manifest_uuid!#].</key>
<key name="job_0076">[ Error ] - Failed to load and parse the install manifest. Details will be found in the logs.</key>
<key name="job_0076">[ Error ] - Failed to load and parse the install manifest. Details will be found in the logs. Exiting, This is a fatal error.</key>
<key name="job_0077">The host name is already: [#!variable!host_name!#], no change needed.</key>
<key name="job_0078">Updating the network configuration for: [#!variable!interface!#].</key>
<key name="job_0079">Disconnected from all database(s). Will reconnect after the network configuration changes have taken effect.</key>
@ -2677,7 +2678,7 @@ Here we will inject 't_0006', which injects 't_0001' which has a variable: [#!st
<key name="warning_0043">[ Warning ] - The install manifest with the UUID: [#!variable!uuid!#] was not found.</key>
<key name="warning_0044">[ Warning ] - The install manifest: [#!variable!name!#] with the UUID: [#!variable!uuid!#] has already been deleted.</key>
<key name="warning_0045">[ Warning ] - The install manifest: [#!variable!name!#] with the UUID: [#!variable!uuid!#] was NOT deleted. The reason may be in the: [#!data!path::log::main!#] log file on this host.</key>
<key name="warning_0046">[ Warning ] - The Install Manifest with the UUID: [#!variable!uuid!#] was not found.</key>
<key name="warning_0046">[ Warning ] - The install manifest with the UUID: [#!variable!uuid!#] was not found.</key>
<key name="warning_0047">[ Warning ] - The password to set for this #!string!brand_0006!# was not set.</key>
<key name="warning_0048">[ Warning ] - The password verification was not set.</key>
<key name="warning_0049">[ Warning ] - The passwords do not match.</key>
@ -2793,6 +2794,7 @@ Read UUID: .... [#!variable!read_uuid!#]
<key name="warning_0118">[ Warning ] - The 'admin' group was created as a system group with the group ID: [#!variable!gid!#].</key>
<key name="warning_0119">[ Warning ] - The 'admin' user was created with the user ID: [#!variable!uid!#].</key>
<key name="warning_0120">[ Warning ] - Timed out waiting for the database: [#!variable!uuid!#] to become available.</key>
<key name="warning_0121">[ Warning ] - The Anvil! with the UUID: [#!variable!uuid!#] was not found. Exiting, will re-run the anvil-join-anvil job again in a few moments.</key>
<!-- The entries below here are not sequential, but use a key to find the entry. -->
<!-- Run 'striker-parse-os-list to find new entries. -->

54
tools/anvil-join-anvil
Unescape View File

@ -1912,19 +1912,20 @@ sub load_job
$machine = "" if not defined $machine;
$manifest_uuid = "" if not defined $manifest_uuid;
$anvil_uuid = "" if not defined $anvil_uuid;
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 3, list => {
machine => $machine,
manifest_uuid => $manifest_uuid,
anvil_uuid => $anvil_uuid,
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 1, list => {
's1:job_data' => $anvil->data->{jobs}{job_data},
's2:machine' => $machine,
's3:manifest_uuid' => $manifest_uuid,
's4:anvil_uuid' => $anvil_uuid,
}});
if ((not $machine) or
(not $manifest_uuid) or
(not $anvil_uuid))
{
# Terminate the job entirely, it's likely an unrecoverable problem.
update_progress($anvil, 100, "job_0092,!!job-uuid!".$anvil->data->{switches}{'job-uuid'}."!!,!!raw!".$anvil->data->{jobs}{job_data}."!!");
$anvil->Log->entry({source => $THIS_FILE, line => __LINE__, 'print' => 1, level => 0, priority => "err", key => "job_0092", variables => {
# This occassionally is hit, but then works when tried again.
update_progress($anvil, 100, "error_0308,!!job-uuid!".$anvil->data->{switches}{'job-uuid'}."!!,!!raw!".$anvil->data->{jobs}{job_data}."!!");
$anvil->Log->entry({source => $THIS_FILE, line => __LINE__, 'print' => 1, level => 0, priority => "err", key => "error_0308", variables => {
'job-uuid' => $anvil->data->{switches}{'job-uuid'},
raw => $anvil->data->{jobs}{job_data},
}});
@ -1941,34 +1942,41 @@ sub load_job
"sys::anvil_uuid" => $anvil->data->{sys}{anvil_uuid},
}});
# Load in the host info and the manifest.
# Load in the host, manifest and anvil data.
$anvil->Database->get_hosts();
$anvil->Database->get_manifests();
$anvil->Database->get_anvils();
# Parse the manifest
if (not exists $anvil->data->{manifests}{manifest_uuid}{$manifest_uuid})
{
$anvil->Log->entry({source => $THIS_FILE, line => __LINE__, 'print' => 1, level => 0, priority => "err", key => "warning_0046", variables => { uuid => $anvil->data->{sys}{manifest_uuid} }});
update_progress($anvil, 0, "warning_0046,!!uuid!".$anvil->data->{sys}{manifest_uuid}."!!");
sleep 10;
$anvil->nice_exit({exit_code => 2});
}
# Parse the manifest
my $problem = $anvil->Striker->load_manifest({manifest_uuid => $anvil->data->{sys}{manifest_uuid}});
$anvil->Log->variables({source => $THIS_FILE, line => __LINE__, level => 3, list => {
manifest_uuid => $anvil->data->{sys}{manifest_uuid},
problem => $problem,
}});
if ($problem)
{
# Something went wrong
$anvil->Log->entry({source => $THIS_FILE, line => __LINE__, 'print' => 1, level => 0, priority => "err", key => "warning_0046", variables => { uuid => $anvil->data->{sys}{manifest_uuid} }});
update_progress($anvil, 0, "job_0076");
sleep 2;
# Something went wrong, fatally. Abort the job.
$anvil->Log->entry({source => $THIS_FILE, line => __LINE__, 'print' => 1, level => 0, priority => "err", key => "job_0076", variables => { uuid => $anvil->data->{sys}{manifest_uuid} }});
update_progress($anvil, 100, "job_0076,!!uuid!".$anvil->data->{sys}{manifest_uuid}."!!");
$anvil->nice_exit({exit_code => 2});
}
# Load the manifest and anvil data.
$anvil->Database->get_anvils();
if ((not exists $anvil->data->{anvils}{anvil_uuid}{$anvil_uuid}) or (not exists $anvil->data->{manifests}{manifest_uuid}{$manifest_uuid}))
# Make sure we have a valid Anvil!
if (not exists $anvil->data->{anvils}{anvil_uuid}{$anvil_uuid})
{
# Terminate the job entirely, it's likely an unrecoverable problem.
update_progress($anvil, 100, "job_0092,!!job-uuid!".$anvil->data->{switches}{'job-uuid'}."!!,!!raw!".$anvil->data->{jobs}{job_data}."!!");
$anvil->Log->entry({source => $THIS_FILE, line => __LINE__, 'print' => 1, level => 2, key => "job_0092", variables => {
'job-uuid' => $anvil->data->{switches}{'job-uuid'},
raw => $anvil->data->{jobs}{job_data},
}});
sleep 2;
# Odd. Error out, the Anvil! might not be loaded yet.
$anvil->Log->entry({source => $THIS_FILE, line => __LINE__, 'print' => 1, level => 0, priority => "err", key => "warning_0121", variables => { uuid => $anvil->data->{sys}{anvil_uuid} }});
update_progress($anvil, 0, "warning_0121,!!uuid!".$anvil->data->{sys}{anvil_uuid}."!!");
sleep 10;
$anvil->nice_exit({exit_code => 5});
}

Write Preview
Loading…
Cancel
Save