On Tue, Jun 25, 2019 at 10:16 PM Yan, Zheng <zyan@xxxxxxxxxx> wrote: > On 6/25/19 9:39 PM, Jeff Layton wrote: > > In any case, I thought blacklisting mostly occurred when clients fail to > > give up their MDS caps. Why would repeated polling create more blacklist > > entries? > > > > mds blacklist client when client is unresponsive. The blacklist entry > stays in osdmap for a day. If client auto reconnect, a laggy client can > keep reconnecting and getting blacklisted. We can address this problem in two directions: 1) MDS simply rejects a client session from a certain IP address if the MDS has been repeatedly blacklisting clients from that IP. If the client never has a new session, there is no need add a new blacklist entry. 2) The client can be smarter and not retry connections if it has been blacklisted ~3 times in the last hour. (Just be sure to log that it is deferring reconnect until later when it thinks the session will be successful.) -- Patrick Donnelly, Ph.D. He / Him / His Senior Software Engineer Red Hat Sunnyvale, CA GPG: 19F28A586F808C2402351B93C3301A3E258DD79D
