Best practice for scratch partition

[Sreejesh_D]

We have ESXi 41 hosts installed on USB flash drives. Recently we had host failures due to APD, unfortuantely we dint have logs for dignosing the issue, since there was no scratch configured.

from the documents its mentioned that there should be a local hard disk (for 4GB vfat) or shared VMFS for confgiuring scractch. Management is not happy in adding disks only for 4 GB volume. And shared storage is not appropraite here since the storage itself failed.

Is there any better way to collect the logs which are generated at the time of failure? can we use SD (4Gb) cards for scratch?

syslog is not an option here coz management network failed at the time when the host was hung.

[peetz]

Having a SAN storage failure and a management network failure at the same time should be a really rare case (I know that this happened to you, but did it happen really simultaneously, in the same minute or even second?).

Most people use a syslog server for collecting system messages. When doing this you need to be aware of that ESXi is very chatty and produces a ridiculously high amount of logs, so plan for sufficient disk space and/or do frequent log rotation and compression on the syslog server.

Having a scratch partition on the SAN is also known to be good practice (see KB1033696), although IMO it also introduces an additional risk (what happens to ESXi when the datastore used for scratch becomes inaccessible?).

Having a scratch partition on a local SD or USB device is not a good idea, because write cycles are very limited on such devices (compared to hard disks) meaning that they will fail quite soon.

Hope this helps

- Andreas