> It also means that sysupgrade would loop forever in case of a
> really
> fatal issue, like when the image is too large for the given mtd
> partition.

maybe we should limit the loop to X times?

> I'd say the mtd util should perform 3-5 consecutive tries in case
> of a
> block erase problem and then finally exit with a non zero code to
> notify
> the caller about the problem.
> 
> The exit code could convey the nature of the problem, e.g. erase
> issue
> or mtd too short; this would allow sysupgrade to take appropriate
> actions without brute-forcing the flash into death.

that sounds very good to me.

a non-reliable sysupgrade is a real showstopper
for the professional world. one of the important
goals of openwrt is to change the bad update-handling
of embedded targets. so a kernel-update from time 
to time should be possible without having to fear.

in our (special) mesh-network case it's a lot more
problematic, because we do everything over the air and
often the devices are nearly unreachable. in our last
update-session we lost 180 routers (but a lot more
didn't have any problems during this upgrade).
(meaning: even 1% fail is too much)

bye, bastian

_______________________________________________
openwrt-devel mailing list
[email protected]
https://lists.openwrt.org/mailman/listinfo/openwrt-devel

Reply via email to