I suggest to use XFS because libAIO still perform many blocking operations with ext4 while approaching to the max depth I/O queue eh under heavy load or with a slow disk
Il ven 30 apr 2021, 21:21 Anton Roskvist (Jira) <[email protected]> ha scritto: > > [ > https://issues.apache.org/jira/browse/ARTEMIS-3272?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17337580#comment-17337580 > ] > > Anton Roskvist commented on ARTEMIS-3272: > ----------------------------------------- > > I have attached the files you requested. I don't really know what happened > to the formatting in the original file, but I made these with windows line > endings, hope they look better. > > Yes, I can sort of reproduce it... it happens several times a day in a > testing environment. So not on demand, but reliably. > > > Remaining AIO issue from ARTEMIS-3084 > > ------------------------------------- > > > > Key: ARTEMIS-3272 > > URL: https://issues.apache.org/jira/browse/ARTEMIS-3272 > > Project: ActiveMQ Artemis > > Issue Type: Bug > > Affects Versions: 2.16.0, 2.17.0 > > Reporter: Anton Roskvist > > Priority: Major > > Attachments: broker.xml, formatted_stack.txt, stack, stack.txt > > > > > > Hi, > > Starting with artemis-2.16.0 I was getting these WARN messages, followed > by a broker shutdown: > > {quote} > > WARN [org.apache.activemq.artemis.core.server] AMQ222010: Critical IO > Error, shutting down the server. > file=AIOSequentialFile:/path/to/artemis-2.16.0/data/journal/activemq-data-2665706.amq, > message=Timeout on close: java.io.IOException: Timeout on close > > {quote} > > They were not that frequent, maybe once per week or so. I read that this > was a known issue that was supposed to be resolved in artemis-2.17.0. > > After upgrading I see this WARN message instead, followed by a thread > dump: > > {quote}WARN [org.apache.activemq.artemis.journal] File > activemq-data-49.amq still has pending IO before closing it > > {quote} > > This does not appear to cause any issues as far as I can tell but it is > printed several times a day so the log is completely cluttered (from the > thread dump). > > I am attaching one of the thread dumps from my logs here: [^stack] > > In this setup I am running a static broker cluster of 5 brokers, each > processing messages at about 40-50 Mbps > > Clients are primarily using OpenWire > > ext4 is used as the filesystem > > Br, > > Anton > > > > -- > This message was sent by Atlassian Jira > (v8.3.4#803005) >
