Scrubs and especially deep-scrubs a

Scrubs and especially deep-scrubs are not even vaguely spread out temporarily. This week I had scrubs and deep-scrubs disabled in one cluster for multiple days while I deployed ten OSD's. The OSD disks are 45x10x3TB, with co-located journals (yes, I know, bad idea), and the cluster is roughly 25% full. When I unset noscrub, nodeep-scrub, a deep scrubbing storm ensued, which caused significant impact to client operations. We observed at least 79 deep scrubs running concurrently, 17% of the OSD's in the cluster. That's too many, and I had to disable scrubs again.

Reference the AMANDA backup system's 'planner' for an approach: full and incremental backups are continually and adaptively spread out over the course of a backup cycle, which maps pretty well to deep-scrubs, scrubs, and osd_scrub_max_interval respectively. Examples are shown below of how clumpy scrubs are even on a long-running cluster, and as a cluster fills up, the more impactful scrub parallelism becomes. 0.80.8+ introduces iopriority settings, which are useful, but scrubs really need to get spread out better.

結果 (中文) 1:

磨砂和特别是深磨砂不甚至隐约散布暂时。这周有磨砂和深磨砂尽管部署了十 OSD 多天在一个群集中禁用。OSD 磁盘是 45x10x3TB，与合用期刊 (是的我知道，好主意)，和群集大约是 25%已满。当我取消设置的 noscrub，nodeep-灌丛，深擦洗风暴随之而来，造成对客户的业务造成重大影响。我们观察到至少 79 深磨砂同时，运行 OSD 在群集中的 17%。这就是太多了，，我不得不再次禁用磨砂。引用阿曼达备份系统 '计划' 的一种方法: 完全备份和增量备份不断和自适应地散布在一个备份周期，哪些映射漂亮好到深磨砂、 磨砂和 osd_scrub_max_interval 的分别，当然。例子如下所示的块状磨砂是即使在长时间运行群集上，如何以及群集已满，变得更有影响力的磨砂膏并行。0.80.8+ 介绍了 iopriority 设置，这些都是很有用，但磨砂真的需要得到更好地传播。

結果 (中文) 3:

磨砂和特别深的灌木，甚至隐约展开暂时。这个星期我有磨砂和深层磨砂残疾人一簇多天，我部署了十个OSD的OSD磁盘45x10x3tb，合作期刊（是的，我知道，坏主意），和集群大致是25%全。当我不noscrub，nodeep擦洗，擦洗的风暴随之而来的深，这对客户运营造成了重大影响。我们观察到至少79个深层磨砂同时运行，对OSD的集群中的17%。太多，我不得不禁用磨砂了。

参考阿曼达备份系统的规划”的方法：完全备份和增量备份是不断地扩散到一个备份周期过程，这地图很深的磨砂，磨砂，和osd_scrub_max_interval分别。例子如下如何块状磨砂甚至在长时间运行的集群，而集群填满，更有效的并行成为擦洗。0.80.8介绍iopriority设置，这是有用的，但磨砂真的需要展开
更好。