docs: rework section on node storage #9652
Conversation
|
|
||
| - cpu/network load per store | ||
| - ranges that are used together often in queries | ||
| - number of active ranges per store |
There was a problem hiding this comment.
Add number of range leases held per store.
| - cpu/network load per store | ||
| - ranges that are used together often in queries | ||
| - number of active ranges per store | ||
| - ranges that already have some overlap (to decrease the chance of data loss) |
There was a problem hiding this comment.
Can you elaborate on this? I'm not sure what it means so it probably needs to be rephrased.
There was a problem hiding this comment.
A less strict way of doing copysets.
I just removed it. It's theoretical and hard to sum up quickly. If we add it, when can update the doc.
| [here](http://docs.basho.com/riak/latest/ops/advanced/backends/leveldb/). | ||
| # Stores and Storage | ||
|
|
||
| Nodes contain one ore more stores. And each of those stores should be on a |
| # Stores and Storage | ||
|
|
||
| Nodes contain one ore more stores. And each of those stores should be on a | ||
| unique disk and contains an instance of RocksDB. And these stores in turn have |
There was a problem hiding this comment.
The second sentence is slightly odd. We create the RocksDB instances. Perhaps: Each store should be placed on a unique disk. Internally, each store contains a single instance of RocksDB with a block cache shared amongst all of the stores in a node.
| Nodes contain one ore more stores. And each of those stores should be on a | ||
| unique disk and contains an instance of RocksDB. And these stores in turn have | ||
| a collection of range replicas. More than one replica for a range will never | ||
| occur on the same store or even the same node. |
| Early on, when a cluster is first initialized, the few default starting ranges | ||
| will only have a single replica, but as soon as other nodes are available they | ||
| will replicate to them until they've reached their desired replication factor, | ||
| the default being 3. Ranges can have different replication factors and when set |
There was a problem hiding this comment.
Mention zone configs?
There was a problem hiding this comment.
Yeah, I've re-written the section to include them.
| the default being 3. Ranges can have different replication factors and when set | ||
| they will up or down replicate to the appropriate number of replicas. Since | ||
| ranges are only created via splits, all replicas for a range split at the same | ||
| time so the replication factors rarely need to be adjusted. |
There was a problem hiding this comment.
I'm not sure how the first part of the sentence connects to the second. Why does the splitting of a range to create new ranges affect the desire to change replication factors?
| To combat this and to spread the overall load across the full cluster, replicas | ||
| will be moved between stores maintaining the desired replication factor. The | ||
| heuristics used to perform this rebalancing are in flux, but they take into | ||
| account a number of different factors: |
There was a problem hiding this comment.
I'd leave out the in flux part and just describe the currently used heuristics and the load-based ones we're considering for the future.
This adds a little bit on repair and rebalancing as well. Part of cockroachdb#9634.
BramGruneir
force-pushed
the
doc-updates2
branch
from
2c0f3f1 to
0a0d79a
Compare
|
Made the requested changes. |
Previously, the `kv/restart/nodes=12` roachtest was unable to pass and was skipped. Following the changes introduced for cockroachdb#9652, the test now passes. This commit enables the `kv/restart/nodes=12` roachtest as a weekly test. Informs: cockroachdb#95159 Release note: None
Previously, the `kv/restart/nodes=12` roachtest was unable to pass and was skipped. Following the changes introduced for cockroachdb#9652, the test now passes. This commit enables the `kv/restart/nodes=12` roachtest as a weekly test. informs: cockroachdb#95159 resolves: cockroachdb#98296 Release note: None
Previously, the `kv/restart/nodes=12` roachtest was unable to pass and was skipped. Following the changes introduced for cockroachdb#9652, the test now passes. This commit enables the `kv/restart/nodes=12` roachtest as a weekly test. resolves: cockroachdb#98296 Release note: None
This adds a little bit on repair and rebalancing as well.
Part of #9634.
This change is