[ https://issues.apache.org/jira/browse/ZOOKEEPER-4788?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17809143#comment-17809143 ]
Enrico Olivelli commented on ZOOKEEPER-4788: -------------------------------------------- Could you please start a discussion on the mailing list ? This discussion deserves more participants and a few people are watching JIRA, it is only our bug tracker. Could you please describe your access pattern? Zookeeper writes must not be in a hot path (with high thoughtput) of an application. As every write is a quorum operation. > high maxlatency > --------------- > > Key: ZOOKEEPER-4788 > URL: https://issues.apache.org/jira/browse/ZOOKEEPER-4788 > Project: ZooKeeper > Issue Type: Bug > Affects Versions: 3.7.1 > Reporter: yangoofy > Priority: Major > > We use Zookeeper as the registration center for microservices. Zookeeper uses > the default configuration. Snapshot600m, with 70w ephemeral nodes. A single > Zookeeper server is 56c128g and xmx16g. The number of client connections is > 8k, and the number of watches is 400w. During the centralized start and stop > time of the client, the maxlatency of the server is as high as 10 seconds. > How can me optimize it? -- This message was sent by Atlassian Jira (v8.20.10#820010)