Ansible deployment state (#17653)

New status flag in the base class to record if Ansible has been executed or not. Use the flag to only to call Ansible deregister when needed. mr_test schedule to use qesap_cleanup in place of peering_destroy Both peering_destroy and qesap_cleanup are doing exactly the same from the point of view of the network peering to the IBSm. qesap_cleanup has cluster destruction after that.
os-autoinst · Sep 25, 2023 · 7c02f5f · 7c02f5f
1 parent fc19dde
commit 7c02f5f
Show file tree

Hide file tree

Showing 20 changed files with 191 additions and 128 deletions.
diff --git a/lib/mr_test_lib.pm b/lib/mr_test_lib.pm
@@ -65,8 +65,10 @@ sub load_mr_tests {
     #   [debug] ||| finished ssh_interactive_end publiccloud
     #   [debug] ||| starting 1_saptune_notes .../mr_test_run.pm
     #   [debug] ||| finished 1_saptune_notes lib
-    loadtest_mr_test('tests/publiccloud/ssh_interactive_end', run_args => $args) if get_var('PUBLIC_CLOUD_SLES4SAP');
-    loadtest_mr_test('tests/sles4sap/publiccloud/peering_destroy', run_args => $args) if get_var('IS_MAINTENANCE');
+    if (get_var('PUBLIC_CLOUD_SLES4SAP')) {
+        loadtest_mr_test('tests/publiccloud/ssh_interactive_end', run_args => $args);
+        loadtest_mr_test('tests/sles4sap/publiccloud/qesap_cleanup', run_args => $args);
+    }
 }
 
 1;
diff --git a/lib/sles4sap_publiccloud.pm b/lib/sles4sap_publiccloud.pm
@@ -50,6 +50,8 @@ our @EXPORT = qw(
   create_instance_data
   deployment_name
   delete_network_peering
+  create_playbook_section_list
+  create_hana_vars_section
 );
 
 =head2 run_cmd
@@ -608,4 +610,58 @@ sub delete_network_peering {
     }
 }
 
+=head2 create_ansible_playbook_list
+
+    Detects HANA/HA scenario from openQA variables and returns a list of ansible playbooks to include
+    in the "ansible: create:" section of config.yaml file.
+
+=cut
+
+sub create_playbook_section_list {
+    my ($ha_enabled) = @_;
+    my @playbook_list;
+
+    # Add registration module as first element - "QESAP_SCC_NO_REGISTER" skips scc registration via ansible
+    push @playbook_list, 'registration.yaml -e reg_code=' . get_required_var('SCC_REGCODE_SLES4SAP') . " -e email_address=''"
+      unless (get_var('QESAP_SCC_NO_REGISTER'));
+
+    # SLES4SAP/HA related playbooks
+    if ($ha_enabled) {
+        push @playbook_list, 'pre-cluster.yaml', 'sap-hana-preconfigure.yaml -e use_sapconf=' . get_required_var('USE_SAPCONF');
+        push @playbook_list, 'cluster_sbd_prep.yaml' if (check_var('FENCING_MECHANISM', 'sbd'));
+        push @playbook_list, qw(
+          sap-hana-storage.yaml
+          sap-hana-download-media.yaml
+          sap-hana-install.yaml
+          sap-hana-system-replication.yaml
+          sap-hana-system-replication-hooks.yaml
+          sap-hana-cluster.yaml
+        );
+    }
+    return (\@playbook_list);
+}
+
+=head2 create_hana_vars_section
+
+    Detects HANA/HA scenario from openQA variables and creates "terraform: variables:" section in config.yaml file.
+
+=cut
+
+sub create_hana_vars_section {
+    my ($ha_enabled) = @_;
+    # Cluster related setup
+    my %hana_vars;
+    if ($ha_enabled == 1) {
+        $hana_vars{sap_hana_install_software_directory} = get_required_var('HANA_MEDIA');
+        $hana_vars{sap_hana_install_master_password} = get_required_var('_HANA_MASTER_PW');
+        $hana_vars{sap_hana_install_sid} = get_required_var('INSTANCE_SID');
+        $hana_vars{sap_hana_install_instance_number} = get_required_var('INSTANCE_ID');
+        $hana_vars{sap_domain} = get_var('SAP_DOMAIN', 'qesap.example.com');
+        $hana_vars{primary_site} = get_var('HANA_PRIMARY_SITE', 'site_a');
+        $hana_vars{secondary_site} = get_var('HANA_SECONDARY_SITE', 'site_b');
+        set_var('SAP_SIDADM', lc(get_var('INSTANCE_SID') . 'adm'));
+    }
+    return (\%hana_vars);
+}
+
 1;
diff --git a/lib/sles4sap_publiccloud_basetest.pm b/lib/sles4sap_publiccloud_basetest.pm
@@ -17,7 +17,7 @@ use qesapdeployment;
 use sles4sap_publiccloud;
 use publiccloud::utils;
 
-our @EXPORT = qw(cleanup);
+our @EXPORT = qw(cleanup import_context);
 
 
 sub cleanup {
@@ -26,17 +26,21 @@ sub cleanup {
     record_info('Cleanup',
         join(' ',
             'cleanup_called:', $self->{cleanup_called} // 'undefined',
-            'network_peering_present:', $self->{network_peering_present} // 'undefined'));
+            'network_peering_present:', $self->{network_peering_present} // 'undefined',
+            'ansible_present:', $self->{ansible_present} // 'undefined'));
     # Do not run destroy if already executed
     return if ($self->{cleanup_called});
     $self->{cleanup_called} = 1;
 
     qesap_upload_logs();
-    delete_network_peering() if ($self->{network_peering_present});
+    if ($self->{network_peering_present}) {
+        delete_network_peering();
+        $self->{network_peering_present} = 0;
+    }
 
     my @cmd_list;
-    # Only run the Ansible deregister if the inventory is present
-    push(@cmd_list, 'ansible') if (!script_run 'test -f ' . qesap_get_inventory(provider => get_required_var('PUBLIC_CLOUD_PROVIDER')));
+    # Only run the Ansible deregister if Ansible has been executed
+    push(@cmd_list, 'ansible') if ($self->{ansible_present});
 
     # Terraform destroy can be executed in any case
     push(@cmd_list, 'terraform');
@@ -49,6 +53,7 @@ sub cleanup {
             if ($cleanup_cmd_rc[0] == 0) {
                 diag(ucfirst($command) . " cleanup attempt # $_ PASSED.");
                 record_info("Clean $command", ucfirst($command) . ' cleanup PASSED.');
+                $self->{ansible_present} = 0 if ($command eq 'ansible');
                 last;
             }
             else {
@@ -65,6 +70,19 @@ sub cleanup {
     record_info('Cleanup finished');
 }
 
+sub import_context {
+    my ($self, $run_args) = @_;
+    $self->{instances} = $run_args->{instances};
+    $self->{network_peering_present} = 1 if ($run_args->{network_peering_present});
+    $self->{ansible_present} = 1 if ($run_args->{ansible_present});
+    record_info('CONTEXT LOG', join(' ',
+            'cleanup_called:', $self->{cleanup_called} // 'undefined',
+            'instances:', $self->{instances} // 'undefined',
+            'network_peering_present:', $self->{network_peering_present} // 'undefined',
+            'ansible_present:', $self->{ansible_present} // 'undefined')
+    );
+}
+
 sub post_fail_hook {
     my ($self) = @_;
     if (get_var('QESAP_NO_CLEANUP_ON_FAILURE')) {

diff --git a/tests/sles4sap/publiccloud/add_server_to_hosts.pm b/tests/sles4sap/publiccloud/add_server_to_hosts.pm
@@ -15,11 +15,9 @@ sub test_flags {
 
 sub run {
     my ($self, $run_args) = @_;
-    my $instances = $run_args->{instances};
-    $self->{network_peering_present} = 1 if ($run_args->{network_peering_present});
-    record_info('CONTEXT LOG', "instances:$instances network_peering_present:$self->{network_peering_present}");
+    $self->import_context($run_args);
 
-    foreach my $instance (@{$instances}) {
+    foreach my $instance (@{$self->{instances}}) {
         next if ($instance->{'instance_id'} !~ m/vmhana/);
         record_info("$instance");
 

diff --git a/tests/sles4sap/publiccloud/check_ibsm_embargoed.pm b/tests/sles4sap/publiccloud/check_ibsm_embargoed.pm
@@ -16,9 +16,8 @@ sub test_flags {
 
 sub run() {
     my ($self, $run_args) = @_;
+    $self->import_context($run_args);
     my $instance = $run_args->{my_instance};
-    $self->{network_peering_present} = 1 if ($run_args->{network_peering_present});
-    record_info('CONTEXT LOG', "instance:$instance network_peering_present:$self->{network_peering_present}");
 
     my @repos = split(/,/, get_var('INCIDENT_REPO'));
     my $count = 0;

diff --git a/tests/sles4sap/publiccloud/cluster_add_repos.pm b/tests/sles4sap/publiccloud/cluster_add_repos.pm
@@ -13,27 +13,24 @@ sub test_flags {
     return {fatal => 1, publiccloud_multi_module => 1};
 }
 
-sub run() {
+sub run {
     my ($self, $run_args) = @_;
-    my $instance = $run_args->{my_instance};
-    $self->{network_peering_present} = 1 if ($run_args->{network_peering_present});
-    record_info('CONTEXT LOG', "instance:$instance network_peering_present:$self->{network_peering_present}");
+    $self->import_context($run_args);
 
     set_var('MAINT_TEST_REPO', get_var('INCIDENT_REPO')) if get_var('INCIDENT_REPO');
-    my $prov = get_required_var('PUBLIC_CLOUD_PROVIDER');
     my @repos = split(/,/, get_var('MAINT_TEST_REPO'));
     my $count = 0;
 
     while (defined(my $maintrepo = shift @repos)) {
         next if $maintrepo =~ /^\s*$/;
-        foreach my $instance (@{$run_args->{instances}}) {
+        foreach my $instance (@{$self->{instances}}) {
             next if ($instance->{'instance_id'} !~ m/vmhana/);
             $instance->run_ssh_command(cmd => "sudo zypper --no-gpg-checks ar -f -n TEST_$count $maintrepo TEST_$count",
                 username => 'cloudadmin');
         }
         $count++;
     }
-    foreach my $instance (@{$run_args->{instances}}) {
+    foreach my $instance (@{$self->{instances}}) {
         next if ($instance->{'instance_id'} !~ m/vmhana/);
         $instance->run_ssh_command(cmd => 'sudo zypper -n ref', username => 'cloudadmin', timeout => 1500);
     }

diff --git a/tests/sles4sap/publiccloud/general_patch_and_reboot.pm b/tests/sles4sap/publiccloud/general_patch_and_reboot.pm
@@ -15,25 +15,25 @@ use testapi;
 use registration;
 use utils;
 use publiccloud::ssh_interactive qw(select_host_console);
-use publiccloud::utils qw(kill_packagekit);
+use publiccloud::utils qw(kill_packagekit is_azure);
 
 sub test_flags {
     return {fatal => 1, publiccloud_multi_module => 1};
 }
 
 sub run {
     my ($self, $run_args) = @_;
-    $self->{network_peering_present} = 1 if ($run_args->{network_peering_present});
+    $self->import_context($run_args);
     select_host_console();    # select console on the host, not the PC instance
 
-    foreach my $instance (@{$run_args->{instances}}) {
+    foreach my $instance (@{$self->{instances}}) {
         next if ($instance->{'instance_id'} !~ m/vmhana/);
         record_info("$instance");
 
         my $remote = '-o ControlMaster=no -o StrictHostKeyChecking=no -o UserKnownHostsFile=/dev/null ' . $instance->username . '@' . $instance->public_ip;
 
         my $cmd_time = time();
-        my $ref_timeout = check_var('PUBLIC_CLOUD_PROVIDER', 'AZURE') ? 3600 : 240;
+        my $ref_timeout = is_azure ? 3600 : 240;
         kill_packagekit($instance);
         $instance->ssh_script_retry("sudo zypper -n --gpg-auto-import-keys ref", timeout => $ref_timeout, retry => 6, delay => 60);
         record_info('zypper ref time', 'The command zypper -n ref took ' . (time() - $cmd_time) . ' seconds.');

diff --git a/tests/sles4sap/publiccloud/hana_sr_schedule_cleanup.pm b/tests/sles4sap/publiccloud/hana_sr_schedule_cleanup.pm
@@ -10,18 +10,25 @@ package hana_sr_schedule_cleanup;
 use strict;
 use warnings FATAL => 'all';
 use base 'sles4sap_publiccloud_basetest';
-use main_common 'loadtest';
 use testapi;
+use main_common 'loadtest';
 
 sub test_flags {
     return {fatal => 1, publiccloud_multi_module => 1};
 }
 
 sub run {
     my ($self, $run_args) = @_;
-    $self->{network_peering_present} = 1 if ($run_args->{network_peering_present});
+
+    # Needed to have peering and ansible state propagated in post_fail_hook
+    $self->import_context($run_args);
 
     record_info("Schedule", "Schedule cleanup job");
+    # A test module only to schedule another test module
+    # It is needed to be sure that qesap_cleanup is executed as last test step
+    # Direct schedule of qesap_cleanup is not possible
+    # due to the fact that hana_sr_schedule_cleanup is scheduled
+    # with other test modules that are also using loadtest
     loadtest('sles4sap/publiccloud/qesap_cleanup', name => "Cleanup resources", run_args => $run_args, @_);
 }
 

diff --git a/tests/sles4sap/publiccloud/hana_sr_schedule_deployment.pm b/tests/sles4sap/publiccloud/hana_sr_schedule_deployment.pm
@@ -10,16 +10,18 @@ package hana_sr_schedule_deployment;
 use strict;
 use warnings FATAL => 'all';
 use base 'sles4sap_publiccloud_basetest';
-use main_common 'loadtest';
 use testapi;
+use main_common 'loadtest';
 
 sub test_flags {
     return {fatal => 1, publiccloud_multi_module => 1};
 }
 
 sub run {
     my ($self, $run_args) = @_;
-    $self->{network_peering_present} = 1 if ($run_args->{network_peering_present});
+
+    # Needed to have peering and ansible state propagated in post_fail_hook
+    $self->import_context($run_args);
 
     if (get_var('QESAP_DEPLOYMENT_IMPORT')) {
         loadtest('sles4sap/publiccloud/qesap_reuse_infra', name => 'prepare_existing_infrastructure', run_args => $run_args, @_);

diff --git a/tests/sles4sap/publiccloud/hana_sr_schedule_primary_tests.pm b/tests/sles4sap/publiccloud/hana_sr_schedule_primary_tests.pm
@@ -13,16 +13,18 @@ package hana_sr_schedule_primary_tests;
 use strict;
 use warnings FATAL => 'all';
 use base 'sles4sap_publiccloud_basetest';
-use main_common 'loadtest';
 use testapi;
+use main_common 'loadtest';
 
 sub test_flags {
     return {fatal => 1, publiccloud_multi_module => 1};
 }
 
 sub run {
     my ($self, $run_args) = @_;
-    $self->{network_peering_present} = 1 if ($run_args->{network_peering_present});
+
+    # Needed to have peering and ansible state propagated in post_fail_hook
+    $self->import_context($run_args);
 
     record_info("Schedule", "Executing tests on master Hana DB");
     # 'HANASR_PRIMARY_ACTIONS' - define to override test flow

diff --git a/tests/sles4sap/publiccloud/hana_sr_schedule_replica_tests.pm b/tests/sles4sap/publiccloud/hana_sr_schedule_replica_tests.pm
@@ -13,16 +13,18 @@ package hana_sr_schedule_replica_tests;
 use strict;
 use warnings FATAL => 'all';
 use base 'sles4sap_publiccloud_basetest';
-use main_common 'loadtest';
 use testapi;
+use main_common 'loadtest';
 
 sub test_flags {
     return {fatal => 1, publiccloud_multi_module => 1};
 }
 
 sub run {
     my ($self, $run_args) = @_;
-    $self->{network_peering_present} = 1 if ($run_args->{network_peering_present});
+
+    # Needed to have peering and ansible state propagated in post_fail_hook
+    $self->import_context($run_args);
 
     record_info("Schedule", "Executing tests on secondary site (replica)");
     # 'HANASR_SECONDARY_ACTIONS' - define to override test flow

diff --git a/tests/sles4sap/publiccloud/hana_sr_takeover.pm b/tests/sles4sap/publiccloud/hana_sr_takeover.pm
@@ -9,8 +9,8 @@ use strict;
 use warnings FATAL => 'all';
 use base 'sles4sap_publiccloud_basetest';
 use testapi;
-use publiccloud::utils;
 use sles4sap_publiccloud;
+use publiccloud::utils;
 use serial_terminal 'select_serial_terminal';
 
 sub test_flags {
@@ -19,8 +19,9 @@ sub test_flags {
 
 sub run {
     my ($self, $run_args) = @_;
-    $self->{network_peering_present} = 1 if ($run_args->{network_peering_present});
-    $self->{instances} = $run_args->{instances};
+
+    # Needed to have peering and ansible state propagated in post_fail_hook
+    $self->import_context($run_args);
 
     select_serial_terminal;
     my $test_name = $self->{name};

diff --git a/tests/sles4sap/publiccloud/hana_sr_test_secondary.pm b/tests/sles4sap/publiccloud/hana_sr_test_secondary.pm
@@ -12,8 +12,8 @@
 use strict;
 use warnings FATAL => 'all';
 use base 'sles4sap_publiccloud_basetest';
-use sles4sap_publiccloud;
 use testapi;
+use sles4sap_publiccloud;
 use serial_terminal 'select_serial_terminal';
 use Time::HiRes 'sleep';
 
@@ -23,8 +23,9 @@ sub test_flags {
 
 sub run {
     my ($self, $run_args) = @_;
-    $self->{network_peering_present} = 1 if ($run_args->{network_peering_present});
-    $self->{instances} = $run_args->{instances};
+
+    # Needed to have peering and ansible state propagated in post_fail_hook
+    $self->import_context($run_args);
     croak('site_b is missing or undefined in run_args') if (!$run_args->{site_b});
 
     my $hana_start_timeout = bmwqemu::scale_timeout(600);