azure-image: fix "Resource Disk" support for swapfile

Since "Resource Disk" is ephemeral, we cannot configure swapfile on machine image building time. So e7e7daa was incorrect, we currently mistakenly using rootfs for swapfile. We need to allocate it on instance startup time, not machine image building time. And also it should executed after "Resource Disk" mounted to /mnt. Fixes scylladb#428 Fixes scylladb#431
syuu1228 · Jan 31, 2023 · b3299c8 · b3299c8
1 parent f6e759c
commit b3299c8
Show file tree

Hide file tree

Showing 3 changed files with 31 additions and 29 deletions.
diff --git a/common/scylla-image-setup.service b/common/scylla-image-setup.service
@@ -2,7 +2,6 @@
 Description=Scylla Cloud Image Setup service
 Before=scylla-server.service
 After=network.target
-ConditionPathExists=!/etc/scylla/machine_image_configured
 
 [Service]
 Type=oneshot

diff --git a/common/scylla_image_setup b/common/scylla_image_setup
@@ -6,33 +6,38 @@
 
 import os
 import sys
-import pathlib
-from lib.scylla_cloud import get_cloud_instance, is_gce, is_redhat_variant
+from pathlib import Path
+from lib.scylla_cloud import get_cloud_instance, is_gce, is_azure, is_redhat_variant
 from subprocess import run
 
 if __name__ == '__main__':
-    # On Ubuntu, we configure CPU scaling while AMI building time
-    if is_redhat_variant():
-        run('/opt/scylladb/scripts/scylla_cpuscaling_setup', shell=True, check=True)
-    cloud_instance = get_cloud_instance()
-    run('/opt/scylladb/scylla-machine-image/scylla_configure.py', shell=True, check=True)
+    if is_azure() and not Path('/mnt/swapfile').exists():
+        Path('/etc/systemd/system/mnt-swapfile.swap').unlink(missing_ok=True)
+        run('/opt/scylladb/scripts/scylla_swap_setup --swap-directory /mnt', shell=True, check=True)
+    machine_image_configured = Path('/etc/scylla/machine_image_configured')
+    if not machine_image_configured.exists():
+        # On Ubuntu, we configure CPU scaling while AMI building time
+        if is_redhat_variant():
+            run('/opt/scylladb/scripts/scylla_cpuscaling_setup', shell=True, check=True)
+        cloud_instance = get_cloud_instance()
+        run('/opt/scylladb/scylla-machine-image/scylla_configure.py', shell=True, check=True)
 
-    run('/opt/scylladb/scripts/scylla_sysconfig_setup --nic eth0 --setup-nic', shell=True, check=True)
-    if os.path.ismount('/var/lib/scylla'):
-        if cloud_instance.is_supported_instance_class():
-            # We run io_setup only when ehpemeral disks are available
-            if is_gce():
-                nr_disks = cloud_instance.nvme_disk_count
-                if nr_disks > 0:
+        run('/opt/scylladb/scripts/scylla_sysconfig_setup --nic eth0 --setup-nic', shell=True, check=True)
+        if os.path.ismount('/var/lib/scylla'):
+            if cloud_instance.is_supported_instance_class():
+                # We run io_setup only when ehpemeral disks are available
+                if is_gce():
+                    nr_disks = cloud_instance.nvme_disk_count
+                    if nr_disks > 0:
+                        cloud_instance.io_setup()
+                else:
                     cloud_instance.io_setup()
-            else:
-                cloud_instance.io_setup()
-            run('systemctl daemon-reload', shell=True, check=True)
-            run('systemctl enable var-lib-systemd-coredump.mount', shell=True, check=True)
-            run('systemctl start var-lib-systemd-coredump.mount', shell=True, check=True)
-            # some distro has fstrim enabled by default, since we are using XFS with online discard, we don't need fstrim
-            run('systemctl is-active -q fstrim.timer && systemctl disable fstrim.timer', shell=True, check=True)
+                run('systemctl daemon-reload', shell=True, check=True)
+                run('systemctl enable var-lib-systemd-coredump.mount', shell=True, check=True)
+                run('systemctl start var-lib-systemd-coredump.mount', shell=True, check=True)
+                # some distro has fstrim enabled by default, since we are using XFS with online discard, we don't need fstrim
+                run('systemctl is-active -q fstrim.timer && systemctl disable fstrim.timer', shell=True, check=True)
 
-    if not os.path.ismount('/var/lib/scylla'):
-        print('Failed to initialize RAID volume!')
-    pathlib.Path('/etc/scylla/machine_image_configured').touch()
+        if not os.path.ismount('/var/lib/scylla'):
+            print('Failed to initialize RAID volume!')
+        machine_image_configured.touch()
diff --git a/packer/scylla_install_image b/packer/scylla_install_image
@@ -114,19 +114,16 @@ if __name__ == '__main__':
 
         setup_opt = '--ntp-domain amazon'
         sysconfig_opt = ''
-        swap_opt = '--swap-directory /'
         kernel_opt = ''
         grub_variable = 'GRUB_CMDLINE_LINUX_DEFAULT'
     elif args.target_cloud == 'gce':
         setup_opt = ''
         sysconfig_opt = '--disable-writeback-cache'
-        swap_opt = '--swap-directory /'
         kernel_opt = ''
         grub_variable = 'GRUB_CMDLINE_LINUX_DEFAULT'
     elif args.target_cloud == 'azure':
         setup_opt = ''
         sysconfig_opt = '--disable-writeback-cache'
-        swap_opt = '--swap-directory /mnt'
         kernel_opt = ' rootdelay=300'
         grub_variable = 'GRUB_CMDLINE_LINUX'
 
@@ -141,7 +138,8 @@ if __name__ == '__main__':
     run('/opt/scylladb/scripts/scylla_cpuscaling_setup --force', shell=True, check=True)
 
     run(f'/opt/scylladb/scripts/scylla_sysconfig_setup --set-clocksource {sysconfig_opt}', shell=True, check=True)
-    run(f'/opt/scylladb/scripts/scylla_swap_setup --swap-size-bytes {half_of_diskfree()} {swap_opt}', shell=True, check=True)
+    if args.target_cloud == 'aws' or args.target_cloud == 'gce':
+        run(f'/opt/scylladb/scripts/scylla_swap_setup --swap-size-bytes {half_of_diskfree()} /', shell=True, check=True)
     run('/opt/scylladb/scripts/scylla_coredump_setup', shell=True, check=True)
     dot_mount = '''
 [Unit]