RGW stores a lot of metadata in the db of OSDs [1], so I would expect to see extensive usage on db device if you store billions of objects through RGW.
Anyway, splitting over should not cause OSDs to reboot, and should still work better than not having dedicated db device, unless the db device itself is overloaded. Have you checked the OSD logs to find the root cause of the reboots? I would also check whether the nvme device is overloaded. [1]: https://docs.ceph.com/en/latest/rados/configuration/bluestore-config-ref/#sizing 发件人: Szabo, Istvan (Agoda)<mailto:istvan.sz...@agoda.com> 发送时间: 2021年9月28日 22:49 收件人: Janne Johansson<mailto:icepic...@gmail.com> 抄送: Ceph Users<mailto:ceph-users@ceph.io> 主题: [ceph-users] Re: Billions of objects upload with bluefs spillover cause osds down? Yes, I understand you but that time you wouldn't have issue because it can fit on the OSD or it would use the space on the osd which is ssd. Or I don't know then what's the secret of storing billions of objects, the osds are not even used 20%, if I calculate as of the current situation, even if I use 1x 2TB nvme for wal+db, if the 15.3TB osd will be at 50%, the nvme would be full again, so I should use for 1x 15.3 TB SSD drive 4TB nvme in front of each of them which doesn't makes sense I guess. Istvan Szabo Senior Infrastructure Engineer --------------------------------------------------- Agoda Services Co., Ltd. e: istvan.sz...@agoda.com --------------------------------------------------- -----Original Message----- From: Janne Johansson <icepic...@gmail.com> Sent: Tuesday, September 28, 2021 1:36 PM To: Szabo, Istvan (Agoda) <istvan.sz...@agoda.com> Cc: Ceph Users <ceph-users@ceph.io> Subject: Re: [ceph-users] Billions of objects upload with bluefs spillover cause osds down? Email received from the internet. If in doubt, don't click any link nor open any attachment ! ________________________________ Den tis 28 sep. 2021 kl 08:15 skrev Szabo, Istvan (Agoda) <istvan.sz...@agoda.com>: > Regarding point 2, how can it spillover if I wouldn’t use db device just > block. It can't but it ACTS like if you had 100% spillover. The act of spilling over is a symptom of db sharing device with data. If you have no dedicated device, then ALL db shares device with data. If you have a db device and it is too small for the whole DB, then PARTS will spill over and you get a notice so you can decide how to act. The spillover is not the problem, it is a symptom of "I have more DB than my db device can hold". No DB device means "immediately I get the same effect as 100% spillover". Think of it as a small glass of water spilling over when you over fill it. If the glass is small, lots of spillover. If the glass is missing, ALL water "spills over". -- May the most significant bit of your life be positive. _______________________________________________ ceph-users mailing list -- ceph-users@ceph.io To unsubscribe send an email to ceph-users-le...@ceph.io _______________________________________________ ceph-users mailing list -- ceph-users@ceph.io To unsubscribe send an email to ceph-users-le...@ceph.io
