MongoDB Jira系列-3.0.6修復(fù)的那些事兒

lyhabc

經(jīng)常有朋友向我咨詢版本選型的問題。

受到老畢同學(xué)的啟發(fā)，E叔決定為大家搬運(yùn)+翻譯+簡單解釋 MongoDB3.0.4+以后各個(gè)版本fixed的比較重要的jira issues。大家可以根據(jù)需求來決定版本選型。

今天是3.0.6篇。

Issues fixed in 3.0.6

以下均是3.0.6中修復(fù)了的jira issues。（比較重要的issues）

由于人肉整理，難免有遺漏，歡迎補(bǔ)充。

SERVER-20248

與連接數(shù)&session有關(guān)的內(nèi)存泄露一則

SERVER-19987

每個(gè)mongo與wt的session都做了cursor cache（此前沒有，易造成內(nèi)存泄露）

SERVER-19751

WT panic halt

SERVER-19673，WT-2007

Wt log cache內(nèi)存泄露。

db.serverStatus().wiredTiger.log['total log buffer size']的返回是WT journal使用的內(nèi)存，但是不受WT cache 設(shè)置的限制，如果特別大了，可能會造成OOM，修復(fù)后將限制其大小

Each log slot has a buffer associated with it. That buffer is grown dynamically until it is large enough to hold the consolidated records.

We recently bumped the number of slots, which may alter how much memory log buffers consume.

We should consider limiting how much memory the log system can use (maybe a log cache size?). Or alternatively limiting the size log buffers can grow to – there is almost certainly a limit to how much memory it makes sense to copy and consolidate into a buffer rather than just writing it directly.

SERVER-19573

WT crash一則

SERVER-19538

SSL 與 Auditing 的issue 一則

SERVER-19522

Capped collection 插入效率的fix

SERVER-19464

aggregate的$sort將不會關(guān)閉connection

The sort stage doesn’t properly close the connection. This causes it to be marked as bad and not returned to the pool which leads to many log lines and connection churn. Potentially gigs per hour of log lines on a small cluster.
轉(zhuǎn)載自：http://www.mongoing.com/jira_3.0.6