As we can see here, there are quite a few differences, which appear to reflect what we're seeing: DEVLINKS on the failing instance doesn't include the by-partuuid path, and we're missing all of the ID_PART_ENTRY_* lines from its udev event.
So looks like the next step is probably to determine why the udev event doesn't contain what it should!
OK, the sda1 events provide a substantial difference:
--- failing.sda1_event 2019-07-26 10:42:50.424486349 -0400 USAGE=filesyste m ENTRY_SCHEME= gpt ENTRY_UUID= 23f70cea- 8174-4426- bc82-462217b972 45 ENTRY_TYPE= 0fc63daf- 8483-4772- 8e79-3d69d8477d e4 ENTRY_NUMBER= 1 ENTRY_OFFSET= 227328 ENTRY_SIZE= 62689247 ENTRY_DISK= 8:0 TYPE_NEW= ext4 /dev/disk/ by-id/scsi- 360022480d7a487 2310ff8d8a6789e ec9-part1 /dev/disk/ cloud/azure_ root-part1 /dev/disk/ by-id/scsi- 14d534654202020 20d7a4872310ff4 1908c3b8d8a6789 eec9-part1 /dev/disk/ by-uuid/ d800ba44- 3054-4cb3- 81f7-dfe32f5e8b 48 /dev/disk/ by-label/ cloudimg- rootfs /dev/disk/ azure/root- part1 /dev/disk/ by-id/wwn- 0x60022480d7a48 72310ff8d8a6789 eec9-part1 /dev/disk/ by-path/ acpi-VMBUS: 01-vmbus- 000000000000889 900000000000000 00-lun- 0-part1 /dev/disk/ by-partuuid/ 23f70cea- 8174-4426- bc82-462217b972 45 /dev/disk/ by-id/wwn- 0x60022480d7a48 72310ff8d8a6789 eec9-part1 /dev/disk/ by-uuid/ d800ba44- 3054-4cb3- 81f7-dfe32f5e8b 48 /dev/disk/ by-label/ cloudimg- rootfs /dev/disk/ by-id/scsi- 14d534654202020 20d7a4872310ff4 1908c3b8d8a6789 eec9-part1 /dev/disk/ azure/root- part1 /dev/disk/ by-id/scsi- 360022480d7a487 2310ff8d8a6789e ec9-part1 /dev/disk/ by-path/ acpi-VMBUS: 01-vmbus- 000000000000889 900000000000000 00-lun- 0-part1 /dev/disk/ cloud/azure_ root-part1
+++ success.sda1_event 2019-07-26 12:02:54.536766079 -0400
@@ -53,9 +53,16 @@
ID_FS_VERSION=1.0
ID_FS_TYPE=ext4
ID_FS_
+ID_PART_
+ID_PART_
+ID_PART_
+ID_PART_
+ID_PART_
+ID_PART_
+ID_PART_
fabric_name=root
.ID_FS_
MAJOR=8
MINOR=1
-DEVLINKS=
+DEVLINKS=
TAGS=:systemd:
As we can see here, there are quite a few differences, which appear to reflect what we're seeing: DEVLINKS on the failing instance doesn't include the by-partuuid path, and we're missing all of the ID_PART_ENTRY_* lines from its udev event.
So looks like the next step is probably to determine why the udev event doesn't contain what it should!