I am using cachefilesd to cache the $HOME folders residing on a ZFS
server mounted over NFS. Cachefilesd consistently used to fail to start
upon boot, and would start instantly given a manual start command. This
behaviour was consistent on Stretch and now on Buster as well, on ca. 10
clients. This evening I was trying to find a way to automate the 'manual
start command' upon boot when I came across the
/etc/rc0.d/K01cachefilesd file. To my surprise, that init file uses
(although commented out) an otherwise undocumented variable $STARTTIME
with the comment that some servers fail to start and the init file can
be instructed to wait for <x> seconds before erroring out.

Setting this variable to the value of 2 seems to have fixed my problem
on all clients involved.

If this variable would be present in the default
/etc/default/cachefilesd config file, I would have played around with
that setting much sooner. It doesn't have to be set (or it can be 0 for
all I care), but just having some form of documentation on this option
could help solve other people's issues with this package.