We are not currently running multiple instances of bitbake against the same sstate cache. We are using PR servers, but currently only local to a build, not shared. We do use packages to update targets.
Below is the content of local.conf (generated by kas).

Cheers

Dan

# meta-custom

PACKAGE_CLASSES ?= "package_deb"
LICENSE_FLAGS_ACCEPTED += "commercial"
# Number of parrallel bitbake tasks
BB_NUMBER_THREADS = "6"
# number of parraellel task run by "make" et al
PARALLEL_MAKE = "-j 4"

# meta-custom-tritium-os
BBFILE_PRIORITY_meta-aws = "11"

# meta-local
PRSERV_HOST = "localhost:0"
# It is recommended to activate "buildhistory" for testing the PR service
INHERIT += "buildhistory"
BUILDHISTORY_COMMIT = "1"

MACHINE ??= "containerx86-64"
DISTRO ??= "tritium3-container"
BBMULTICONFIG ?= ""

On Fri, 9 Dec 2022 at 16:11, Richard Purdie <richard.purdie@linuxfoundation.org> wrote:
On Fri, 2022-12-09 at 16:02 +0000, Daniel Squires wrote:
> Hi!
> We're using kirkstone, though not the very latest at present. 
> Every now and then a a number of packages will unexpectedly give the
> error re package versions going backwards and teh version may have
> done something like go from git5-######-r0.1 to git3-#######-r0.0.
> We always see this if we've been using devtool for a given recipe and
> then do a devtool reset, it complains about going back from git999 to
> e.g. git5 but this makes sense and we just run the build again and
> forget about it. However sometimes this just seems to happen randomly
> for a given recipe when it picks up a new commit from our repos and
> the git autoinc number will have gone back a few.
> Does anybody know why and if it is a bug or an be fixed?

We probably need to know a bit more about how you're building things.
Do you have a pool of workers building against a common sstate?

Are you using hash equivalence? Are you using a PR server?

I ask the questions as the answer is probably somewhere in the way data
is being shared/reused between builds.

If you're not using package management on device to upgrade them, it
probably isn't a big issue but it is worth understanding the sstate
reuse in this context.

Cheers,

Richard