Comments (11)
I've got the matrix invite and messaged you, and see no other messages from you. Not sure if you're off somewhere or my matrix server is not happy
from ceph-balancer.
Wohoo I've fixed this in d92c2fc ! Thanks for your perfect reproduction cluster state file @dimm0 😄
from ceph-balancer.
That's bad. And it means my crush validation is somehow wrong. I guess that's also why #4 occurs, and the sanity check fires (with the same reason - i've never tested the script on a cluster having racks, instead of just hosts).
It's really hard debugging without a working cluster currently, I should add a feature where it collects all the json data so you may send it to me and one can test/fix it completely offline.
Also here, for now, it would be helpful if you ran with maximum logging, and send me the output of the whole decision process of one move that'll be invalid, and one of a move that is valid.
from ceph-balancer.
Okay, sorry I should read both your responses before replying :-).
Currently we use cluster in a way that 1 rack = 1 host, but we are currently preparing to extend cluster which will result in multiple hosts within single rack so I cannot really change CRUSH topology right now.
However I think I can send you all necessary JSON outputs which are collected in the beginning of script. I just probably prefer not to attach them publicly to the issue directly. Can we find some more private channel to deliver them? E.g. email or PM somewhere?
from ceph-balancer.
I've now added a feature to export the state bundle :)
Please create it with ./placementoptimizer.py gather outputfile.xz
(you can reach me in matrix or via mail - jj -at- sft lol
:) Ideally upload the file somewhere and send me a link (mail's not suited for blobs..)
from ceph-balancer.
Same here, almost all moves are not allowed in a cluster with datacenter failure domain
from ceph-balancer.
Are running the latest version?
Can you send me your state dump, like i described previously, please? Also, for a given cluster state dump, please send me generated moves that ceph does not accept, but which the balancer produced. That would mean i can tinker with the code and try to figure out the problem.
from ceph-balancer.
Yes, the most recent.
I had to do a couple changes, since json is giving a wrong type for replicated pools (I'm using rook, it might've done something weird in config):
diff --git a/placementoptimizer.py b/placementoptimizer.py
index 91a37ae..25a64e6 100755
--- a/placementoptimizer.py
+++ b/placementoptimizer.py
@@ -320,17 +320,18 @@ for pool in CLUSTER_STATE["df_dump"]["pools"]:
for pool in CLUSTER_STATE["pool_dump"]:
+
id = pool["pool_id"]
ec_profile = pool["erasure_code_profile"]
pg_shard_size_avg = pools[id]["stored"] / pools[id]["pg_num"]
- if ec_profile:
+ if ec_profile and ec_profiles.get(ec_profile):
pg_shard_size_avg /= ec_profiles[ec_profile]["data_chunks"]
pools[id].update({
"erasure_code_profile": ec_profile,
- "repl_type": "ec" if ec_profile else "repl",
+ "repl_type": "ec" if ec_profile and ec_profiles.get(ec_profile) else "repl",
"pg_shard_size_avg": pg_shard_size_avg
})
@@ -678,7 +679,7 @@ def get_pg_shardsize(pgid):
shard_size += pg_stats['num_omap_bytes']
ec_profile = pg_ec_profile(pgid)
- if ec_profile:
+ if ec_profile and ec_profiles.get(ec_profile):
shard_size /= ec_profiles[ec_profile]["data_chunks"]
# omap is not supported on EC pools (yet)
# when it is, check how the omap data is spread (replica or also ec?)
[root@k8s-haosu-14 ~]# ./placementoptimizer.py -v balance --max-pg-moves 10
[2022-03-15 17:13:26,040] gathering cluster state via ceph api...
[2022-03-15 17:13:31,988] running pg balancer
[2022-03-15 17:13:32,023] current OSD fill rate per crushclasses:
[2022-03-15 17:13:32,024] hdd: average=51.24%, median=50.80%, without_placement_constraints=58.60%
[2022-03-15 17:13:32,024] cluster variance for crushclasses:
[2022-03-15 17:13:32,024] hdd: 54.884
[2022-03-15 17:13:32,025] min osd.26 37.306%
[2022-03-15 17:13:32,025] max osd.34 68.550%
[2022-03-15 17:13:32,033] SAVE move 1.345 osd.34 => osd.26
[2022-03-15 17:13:32,033] props: size=81.0G remapped=False upmaps=0
[2022-03-15 17:13:32,033] => variance new=54.63124092496721 < 54.884087241088345=old
[2022-03-15 17:13:32,034] new min osd.19 37.488%
[2022-03-15 17:13:32,034] max osd.29 68.065%
[2022-03-15 17:13:32,034] new cluster variance:
[2022-03-15 17:13:32,034] hdd: 54.631
[2022-03-15 17:13:32,043] SAVE move 1.21b osd.29 => osd.19
[2022-03-15 17:13:32,043] props: size=81.0G remapped=False upmaps=0
[2022-03-15 17:13:32,043] => variance new=54.384077250915475 < 54.63124092496721=old
[2022-03-15 17:13:32,043] new min osd.170 38.010%
[2022-03-15 17:13:32,043] max osd.34 67.683%
[2022-03-15 17:13:32,043] new cluster variance:
[2022-03-15 17:13:32,043] hdd: 54.384
[2022-03-15 17:13:32,052] SAVE move 1.10c osd.34 => osd.170
[2022-03-15 17:13:32,052] props: size=80.9G remapped=False upmaps=0
[2022-03-15 17:13:32,053] => variance new=54.161715943167025 < 54.384077250915475=old
[2022-03-15 17:13:32,053] new min osd.26 38.170%
[2022-03-15 17:13:32,053] max osd.29 67.199%
[2022-03-15 17:13:32,053] new cluster variance:
[2022-03-15 17:13:32,053] hdd: 54.162
[2022-03-15 17:13:32,062] SAVE move 1.75 osd.29 => osd.26
[2022-03-15 17:13:32,062] props: size=80.2G remapped=False upmaps=0
[2022-03-15 17:13:32,062] => variance new=53.9295038278412 < 54.161715943167025=old
[2022-03-15 17:13:32,062] new min osd.133 38.245%
[2022-03-15 17:13:32,062] max osd.34 66.817%
[2022-03-15 17:13:32,062] new cluster variance:
[2022-03-15 17:13:32,062] hdd: 53.930
[2022-03-15 17:13:32,071] SAVE move 1.1a0 osd.34 => osd.133
[2022-03-15 17:13:32,071] props: size=80.4G remapped=False upmaps=0
[2022-03-15 17:13:32,071] => variance new=53.71749631647796 < 53.9295038278412=old
[2022-03-15 17:13:32,071] new min osd.122 38.325%
[2022-03-15 17:13:32,071] max osd.29 66.341%
[2022-03-15 17:13:32,071] new cluster variance:
[2022-03-15 17:13:32,071] hdd: 53.717
[2022-03-15 17:13:32,081] SAVE move 1.38f osd.29 => osd.122
[2022-03-15 17:13:32,081] props: size=80.2G remapped=False upmaps=0
[2022-03-15 17:13:32,082] => variance new=53.510389448720424 < 53.71749631647796=old
[2022-03-15 17:13:32,082] new min osd.19 38.353%
[2022-03-15 17:13:32,082] max osd.34 65.956%
[2022-03-15 17:13:32,082] new cluster variance:
[2022-03-15 17:13:32,082] hdd: 53.510
[2022-03-15 17:13:32,090] SAVE move 1.fd osd.34 => osd.19
[2022-03-15 17:13:32,091] props: size=80.0G remapped=False upmaps=0
[2022-03-15 17:13:32,091] => variance new=53.290570485934 < 53.510389448720424=old
[2022-03-15 17:13:32,091] new min osd.70 38.369%
[2022-03-15 17:13:32,091] max osd.29 65.484%
[2022-03-15 17:13:32,091] new cluster variance:
[2022-03-15 17:13:32,091] hdd: 53.291
[2022-03-15 17:13:32,100] SAVE move 1.1e8 osd.29 => osd.70
[2022-03-15 17:13:32,100] props: size=80.2G remapped=False upmaps=0
[2022-03-15 17:13:32,100] => variance new=53.09083806042146 < 53.290570485934=old
[2022-03-15 17:13:32,100] new min osd.130 38.645%
[2022-03-15 17:13:32,100] max osd.53 65.476%
[2022-03-15 17:13:32,100] new cluster variance:
[2022-03-15 17:13:32,100] hdd: 53.091
[2022-03-15 17:13:32,109] SAVE move 1.11e osd.53 => osd.130
[2022-03-15 17:13:32,109] props: size=80.5G remapped=False upmaps=0
[2022-03-15 17:13:32,109] => variance new=52.892366525271036 < 53.09083806042146=old
[2022-03-15 17:13:32,109] new min osd.170 38.730%
[2022-03-15 17:13:32,109] max osd.34 65.101%
[2022-03-15 17:13:32,109] new cluster variance:
[2022-03-15 17:13:32,109] hdd: 52.892
[2022-03-15 17:13:32,118] SAVE move 28.6de osd.34 => osd.170
[2022-03-15 17:13:32,118] props: size=48.1G remapped=False upmaps=0
[2022-03-15 17:13:32,118] => variance new=52.77434199993548 < 52.89236652527104=old
[2022-03-15 17:13:32,118] new min osd.140 38.759%
[2022-03-15 17:13:32,118] max osd.66 65.037%
[2022-03-15 17:13:32,118] new cluster variance:
[2022-03-15 17:13:32,118] hdd: 52.774
[2022-03-15 17:13:32,118] enough remaps found
[2022-03-15 17:13:32,118] --------------------------------------------------------------------------------
[2022-03-15 17:13:32,118] generated 10 remaps.
[2022-03-15 17:13:32,118] total movement size: 772.4G.
[2022-03-15 17:13:32,118] --------------------------------------------------------------------------------
[2022-03-15 17:13:32,119] old cluster variance per crushclass:
[2022-03-15 17:13:32,119] hdd: 54.884
[2022-03-15 17:13:32,119] old min osd.26 37.306%
[2022-03-15 17:13:32,119] old max osd.34 68.550%
[2022-03-15 17:13:32,119] --------------------------------------------------------------------------------
[2022-03-15 17:13:32,119] new min osd.140 38.759%
[2022-03-15 17:13:32,119] new max osd.66 65.037%
[2022-03-15 17:13:32,119] new cluster variance:
[2022-03-15 17:13:32,119] hdd: 52.774
[2022-03-15 17:13:32,119] --------------------------------------------------------------------------------
ceph osd pg-upmap-items 1.345 34 26
ceph osd pg-upmap-items 1.21b 29 19
ceph osd pg-upmap-items 1.10c 34 170
ceph osd pg-upmap-items 1.75 29 26
ceph osd pg-upmap-items 1.1a0 34 133
ceph osd pg-upmap-items 1.38f 29 122
ceph osd pg-upmap-items 1.fd 34 19
ceph osd pg-upmap-items 1.1e8 29 70
ceph osd pg-upmap-items 1.11e 53 130
ceph osd pg-upmap-items 28.6de 34 170
1.345 20898 0 0 0 0 86954708992 296 27 3386 3386 active+clean 2022-03-14T05:26:58.567934+0000 1908585'13965079 1908585:39935681 [99,34,104] 99 [99,34,104]
I have the datacenter failure domain
1.345 34->26 will move it from datacenter ucsb
to datacenter ucsd
which is also the datacenter of osd.99 (in fact the same node even)
from ceph-balancer.
I'm @dimm:matrix.nrp-nautilus.io on matrix
from ceph-balancer.
Oh, I just read in another ticket that you only support hosts for now. I guess that's the issue then
from ceph-balancer.
It seems that it is the case, but I do want to fix the bugs so we can use it to balance arbitrary crush trees :) I've contacted you on matrix!
from ceph-balancer.
Related Issues (20)
- Fatal error in get_osd_candidates HOT 5
- Do Not Skip the Entire Pool When Balancing PGs HOT 5
- IndexError: list index out of range HOT 3
- can't run it with python < 3.9 HOT 1
- Add ability to handle invoking the Ceph client via Cephadm? HOT 1
- no trace found for 2147483647 in default~hdd HOT 6
- JSONDecodeError in Reef HOT 4
- Please consider type datacenter as a root bucket HOT 1
- JSON TypeError on show command when json output is used. HOT 4
- ./placementoptimizer.py showremapped balks at cluster topology changes HOT 3
- Feature request: take bluestore_min_alloc_size into account HOT 3
- KeyError: 66 (in used += self.cluster.osd_transfer_remainings[osdid]) HOT 5
- ZeroDivisionError: division by zero (osd_objs_acting) HOT 3
- RuntimeError: pg 18.6a to be moved to osd.117 is misplaced with -198781.0<0 objects already transferred
- ./placementoptimizer.py showremapped --by-osd throws KeyError if there is recovery
- key error - on healthy cluster
- it's breaking CRUSH rule HOT 1
- Sorting in heterogeneous clusters
- OSD utilization is very different from ceph osd df HOT 2
- crush root X not known HOT 8
Recommend Projects
-
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
-
Vue.js
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
-
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
-
TensorFlow
An Open Source Machine Learning Framework for Everyone
-
Django
The Web framework for perfectionists with deadlines.
-
Laravel
A PHP framework for web artisans
-
D3
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
-
Recommend Topics
-
javascript
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
-
web
Some thing interesting about web. New door for the world.
-
server
A server is a program made to process requests and deliver data to clients.
-
Machine learning
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
-
Visualization
Some thing interesting about visualization, use data art
-
Game
Some thing interesting about game, make everyone happy.
Recommend Org
-
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
-
Microsoft
Open source projects and samples from Microsoft.
-
Google
Google ❤️ Open Source for everyone.
-
Alibaba
Alibaba Open Source for everyone
-
D3
Data-Driven Documents codes.
-
Tencent
China tencent open source team.
from ceph-balancer.