search

MetPX / Sundew

WMO-GTS switch on TCP/IP protocols (largely obsolete, legacy code.)
GNU General Public License v2.0
5 stars 3 forks source link

px "graceful" ? gentler restart? #14

Open petersilva opened 3 years ago

petersilva commented 3 years ago

when reloading/restarting px, with hundreds of senders and dozens of receivers, each one reads the entire routing table into memory. the load on the system goes very high >100 for a minute or two. This can cause forwarding delays.

otoh, if the configurations are not initiallized, that also causes forwarding delays. anyways... ideal would be to have component restart in small batches over a few minutes to avoid the huge spike in load.

petersilva commented 2 years ago

@junhu3 @kurt2x @habilinour .... we've wanted something like this for ages, to stop cpu load spiking through the roof on px restart... this implementation just spreads it out, so it takes... something like 10 minutes to run, instead of launching them all in a few seconds.

Should I just put it in the release? then if you don't like how it works, just don't use 'graceful'...

junhu3 commented 2 years ago

It's worthy to try. To restart px, I usually do a for i in *conf,.... I think px graceful restart is more practical.