Merge pull request #7834 from VersatusHPC/fix/ubuntu-subiquity-ppc64le-grub-ieee1275

fix(xcat-core): the Ubuntu ppc64le diskful install stops without grub-ieee1275

xCAT-server/lib/perl/xCAT/Template.pm

@@ -1850,6 +1850,18 @@ sub ubuntu_subiquity_apt_mirror
 # The key curtin names in the Deb822 source it writes for the primary apt mirror on 24.04 and later.
 my $UBUNTU_ARCHIVE_KEYRING = '/usr/share/keyrings/ubuntu-archive-keyring.gpg';
 
+# The apt lines that keep the recommended packages out of the install. ospkgs installs without
+# them and the installer matches it. ppc64el is the exception: curtin installs a bootloader
+# package of its own for UEFI and for s390x only, and on a PReP machine install_grub runs
+# "dpkg-reconfigure grub-ieee1275" on a package the Ubuntu kernel image only recommends.
+sub ubuntu_subiquity_no_recommends_lines
+{
+    my ($osarch) = @_;
+
+    return () if defined($osarch) && $osarch =~ /^ppc64/i;
+    return (q(    conf: 'APT::Install-Recommends "false";'));
+}
+
 sub ubuntu_subiquity_apt_config
 {
     my ($media_dir, $osarch, $pkgdirs) = @_;
@@ -1877,7 +1889,7 @@ sub ubuntu_subiquity_apt_config
             '  apt:',
             '    preserve_sources_list: false',
             '    geoip: false',
-            q(    conf: 'APT::Install-Recommends "false";'),
+            ubuntu_subiquity_no_recommends_lines($osarch),
             '    mirror-selection:',
             '      primary:',
             "      - uri: $online_mirror",
@@ -1916,7 +1928,7 @@ sub ubuntu_subiquity_apt_config
         '    preserve_sources_list: false',
         '    fallback: offline-install',
         '    geoip: false',
-        q(    conf: 'APT::Install-Recommends "false";'),
+        ubuntu_subiquity_no_recommends_lines($osarch),
         '    disable_suites:',
         '      - updates',
         '      - backports',

xCAT-server/share/xcat/install/ubuntu/compute.subiquity.tmpl

@@ -127,4 +127,11 @@ autoinstall:
     # monitor that died and never came back is #7759.
     - ['bash', '-c', 'xm=#XCATVAR:XCATMASTER#; port="#TABLEBLANKOKAY:site:key=xcatiport:value#"; [ -n "$port" ] || port=3002; ok=0; for i in 1 2 3 4 5; do if exec 3<>/dev/tcp/$xm/$port; then if read -r -t 10 hello <&3 && [ "$hello" = "ready" ]; then printf "next\n" >&3; if read -r -t 10 ack <&3 && [ "$ack" = "done" ]; then ok=1; fi; fi; exec 3>&- 3<&-; [ "$ok" = 1 ] && break; fi; sleep 5; done; if [ "$ok" != 1 ]; then echo "xcat: FAILED to flip $(hostname) to local-disk boot via $xm:$port; the node will PXE back into the installer" >>/target/var/log/xcat/xcat.log; fi; exit 0']
   error-commands:
-    - tar -c --ignore-failed-read --transform='s/^/#HOSTNAME#-logs\//' /var/crash /var/log/installer /tmp/pre-install.log /autoinstall.yaml 2>/dev/null |nc -l 8080
+    # Subiquity waits for every error command to return. "nc -l 8080" waits for a collector, which
+    # an unattended install does not have. Keep the archive on the installer, and print the end of
+    # the curtin log to the console, which the management node records.
+    #
+    # Both destinations are read from the environment so a test can point them at its own
+    # directory. The installer sets neither, so an install writes where it always did.
+    - ['sh', '-c', 'tar -c --ignore-failed-read --transform="s/^/#HOSTNAME#-logs\//" /var/crash /var/log/installer /tmp/pre-install.log /autoinstall.yaml >"${XCAT_ERROR_ARCHIVE:-/run/#HOSTNAME#-logs.tar}" 2>/dev/null; exit 0']
+    - ['sh', '-c', 'tail -n 80 /var/log/installer/curtin-install.log >"${XCAT_ERROR_CONSOLE:-/dev/console}" 2>/dev/null; exit 0']

xCAT-test/bats/ubuntu_subiquity_error_commands.bats

@@ -0,0 +1,92 @@
+#!/usr/bin/env bats
+#
+# Subiquity waits for every error command to return before it reports the failure. An error
+# command that waits for a collector an unattended install does not have therefore holds the
+# node until the provisioning timeout.
+#
+# Run the error commands, with the programs that would reach the host or the network replaced,
+# and check that they return and that they write the end of the curtin log where the caller
+# points.
+
+load 'helpers/shell_source'
+
+setup()
+{
+    TEMPLATE="$(repo_path 'xCAT-server/share/xcat/install/ubuntu/compute.subiquity.tmpl')"
+    [ -r "$TEMPLATE" ] || skip "$TEMPLATE is required"
+    export TEMPLATE
+}
+
+# One command per list item of the error-commands block. The list form ['sh', '-c', '...']
+# carries the command in its last element; a plain item is the command itself.
+error_commands()
+{
+    awk '
+        /^  error-commands:$/ { copy = 1; next }
+        copy && !/^    [-#]/ { exit }
+        copy && /^    - / {
+            found = 1
+            line = substr($0, 7)
+            if (match(line, /^\[[^]]*, .-c., .*\]$/)) {
+                sub(/^\[[^]]*, .-c., ./, "", line)
+                sub(/.\]$/, "", line)
+            }
+            print line
+        }
+        END { if (!found) exit 1 }
+    ' "$TEMPLATE"
+}
+
+# bash resolves a function ahead of PATH, so the commands run as written while nothing reaches
+# the host or the network. The nc shadow waits the way a listener with no collector waits.
+#
+# Shadowing a command does not stop the shell opening the file the command redirects into: the
+# redirection is the shell's, and it happens whether or not tar runs. So the archive path is
+# taken from the environment too, and pointed inside the test's own directory. Without that,
+# this test would truncate /run/<node>-logs.tar on the host running it, and CI runs as root.
+run_error_commands()
+{
+    local script="${BATS_TEST_TMPDIR}/error-commands.sh" command
+
+    : >"$script"
+    {
+        printf "export XCAT_ERROR_CONSOLE='%s'\n" "$CONSOLE"
+        printf "export XCAT_ERROR_ARCHIVE='%s'\n" "$ARCHIVE"
+        printf 'nc() { sleep 300; }\n'
+        printf 'tar() { echo XCAT_LOGS_ARCHIVE; }\n'
+        printf 'tail() { echo XCAT_CURTIN_LOG_TAIL; }\n'
+    } >>"$script"
+
+    while IFS= read -r command; do
+        # Subiquity runs each error command on its own, so a command that ends in "exit 0"
+        # must not end the others.
+        printf '( %s )\n' "${command//\#HOSTNAME\#/testnode}" >>"$script"
+    done < <(error_commands)
+
+    timeout 10 bash "$script"
+}
+
+@test "the error commands return instead of waiting for someone to collect the logs" {
+    [ -n "$(error_commands)" ]
+    CONSOLE="${BATS_TEST_TMPDIR}/console"
+    ARCHIVE="${BATS_TEST_TMPDIR}/logs.tar"
+
+    run run_error_commands
+    [ "$status" -ne 124 ]
+    grep -q XCAT_CURTIN_LOG_TAIL "$CONSOLE"
+}
+
+@test "the log archive is written where the caller points, and nowhere else" {
+    CONSOLE="${BATS_TEST_TMPDIR}/console"
+    ARCHIVE="${BATS_TEST_TMPDIR}/logs.tar"
+
+    run run_error_commands
+    [ "$status" -ne 124 ]
+
+    # The archive command redirects into the path it was given, so its output IS the proof
+    # that the redirection went there and not to the default under /run. There is nothing to
+    # add by looking at the default path: whether /run/testnode-logs.tar exists says nothing
+    # about this test, because anything else may have created it, and its absence would be
+    # just as true had the override never worked.
+    grep -q XCAT_LOGS_ARCHIVE "$ARCHIVE"
+}

xCAT-test/unit/ubuntu_subiquity_pkglist.t

@@ -193,8 +193,21 @@ is( $render->( $pkglist, environvar => 'ACCEPT_EULA=Y' ), $without,
     };
     like( $apt_render->(), qr{URIs: http://mirror\.example/ubuntu}, 'the pkgdir mirror joins the installer sources' );
     like( $apt_render->(), qr{^    conf: 'APT::Install-Recommends "false";'$}m, 'the installer installs without recommended packages, as ospkgs does' );
+    # On a PReP machine curtin installs no bootloader package of its own: install_missing_packages
+    # covers UEFI and s390x only, and install_grub then runs "dpkg-reconfigure grub-ieee1275".
+    # The Ubuntu kernel image recommends grub-ieee1275, so the package reaches the target only
+    # when the installer keeps recommended packages.
+    unlike( $apt_render->( osarch => 'ppc64el' ), qr{APT::Install-Recommends},
+        'except on ppc64el, where the bootloader the installer configures is a recommended package' );
     unlike( $apt_render->( environvar => 'http_proxy=http://proxy.example:3128' ), qr{mirror\.example|xcat-pkgdir},
         'but not for an osimage with environvar, whose mirrors may need those variables' );
+
+    # The offline configuration installs from the media alone and carries the same setting.
+    local *xCAT::Template::ubuntu_subiquity_apt_mirror = sub { '' };
+    like( $apt_render->(), qr{^    conf: 'APT::Install-Recommends "false";'$}m,
+        'an offline install drops the recommended packages as well' );
+    unlike( $apt_render->( osarch => 'ppc64el' ), qr{APT::Install-Recommends},
+        'and keeps them on ppc64el for the same reason' );
 }
 
 done_testing();