Joe Maimon wrote:
points remain.
- BGP full tables loading/kernel-route syncing can easily lock up the processes to the point where they dont respond properly, and more problematic, watchfrr kills them. Rinse, repeat. This hidden gem looks like it might work crudely, but effectively.
zebra zapi-packets 250 Initial tests are promising. I can run out of memory on the tiny VM before processes lock up or slow thread to death, so far so good.
In other words, it may not be enough to log SLOW THREAD: task zserv_process_messages (55f1e2bb0750) ran for 7819ms (cpu time 253ms) but actually implement some sort of time based thread yielding right in middle of large batch operations, such as loading 800000 routes or so.
- zebra protocol import route-map is strangely one sided. It would be ever so much more powerful to control both what protocols export to zebra and what they import from zebra, as well as allowing protocols to distinguish the source of the route from zebra.
_______________________________________________ frog mailing list frog@lists.frrouting.org https://lists.frrouting.org/listinfo/frog