On Fri, Oct 7, 2016 at 3:09 PM, Felix Fietkau <n...@nbd.name> wrote: > On 2016-10-07 13:57, Zefir Kurtisi wrote: >> In ubus_cli_wait_for() there is a critical section between >> initially checking for the requested services and the >> following handling of 'ubus.object.add' events. >> >> In our system we let procd (re)start services and synchronize >> inter-service dependencies by using 'ubus wait_for' in the >> initscripts' service_started() functions. There we observe >> that 'wait_for' randomly is waiting for the full timeout >> and returning UBUS_STATUS_TIMEOUT, even if the service it >> is waiting for is already up and running. >> >> This happens when the service is started in the critical >> section mentioned above. This commit adds periodic lookup >> for the requested services while waiting for the 'add' event >> and with that fixes the observed failure. >> >> Signed-off-by: Zefir Kurtisi <zefir.kurt...@neratec.com> > Instead of introducing yet another timer, wouldn't it also be possible > to close this race window by registering the event handler before > attempting the lookup? > > - Felix > > _______________________________________________ > Lede-dev mailing list > Lede-dev@lists.infradead.org > http://lists.infradead.org/mailman/listinfo/lede-dev
I've also seen this race. I tried something like this: https://github.com/commodo/ubus/commit/8c3986caaa7cd2c12f2b8907ceea54c5bdce3bd2 But never got around to doing much testing to see if the race goes away completely. So, I never pushed it upstream. @Zefir, maybe you could try it ? Thanks Alex _______________________________________________ Lede-dev mailing list Lede-dev@lists.infradead.org http://lists.infradead.org/mailman/listinfo/lede-dev
