Hi everyone, I've hit a bug with using CPG, and maybe not really related with CPG. While i'm developping futur application based on CPG, i've stressed service, and failed too easily.
Failure happen when CPG message are too big, and some thread are consuming some CPU. When the case happen, corosync failed to send message, and show : [TOTEM ] The token was lost in the OPERATIONAL state. And loop on this while executing test. To make thing easier to test, i've modified cpgbench to : - add a "listener" mode (we just listening, no sending message) - add options to configure start size of message, and grow size - add option to launch background thread who consume some CPU To reproduce bug, you need 2 nodes. One the first, start with: ./cpgbench -l On the second : ./cpgbench -t 10 -s 100000 (start with 10 cpu consumer, and set cpg message size to 100Ko). Maybe you need to increase value if you have a big cpu :) (this failed on CPU: VIA Samuel 2 (400.91-MHz 686-class CPU)) Is someone can help us ? Regards, -- Mathieu Virbel NETASQ - We secure IT 3 rue Archimède 59650 Villeneuve d'Ascq France _______________________________________________ Openais mailing list [email protected] https://lists.linux-foundation.org/mailman/listinfo/openais
