[jira] [Commented] (HDDS-3167) Support different replication factor for CLOSED Container
[ https://issues.apache.org/jira/browse/HDDS-3167?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17102568#comment-17102568 ] Marton Elek commented on HDDS-3167: --- I am planning to pick it up again as it can be related to EC / random write. Will ping you with the draft. > Support different replication factor for CLOSED Container > - > > Key: HDDS-3167 > URL: https://issues.apache.org/jira/browse/HDDS-3167 > Project: Hadoop Distributed Data Store > Issue Type: Improvement >Reporter: Sammi Chen >Priority: Major > > For open container, it's replication is decided by Ratis pipeline number > size. For closed container, this is no such constrain. We can support more > replication factor cases, such as 2 factor, to save storage for some cold > data. -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: ozone-issues-unsubscr...@hadoop.apache.org For additional commands, e-mail: ozone-issues-h...@hadoop.apache.org
[jira] [Commented] (HDDS-3167) Support different replication factor for CLOSED Container
[ https://issues.apache.org/jira/browse/HDDS-3167?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17060843#comment-17060843 ] Sammi Chen commented on HDDS-3167: -- Sure, it's a good idea. It seems https://hackmd.io/ is not accessable here. Google docs works for me. > Support different replication factor for CLOSED Container > - > > Key: HDDS-3167 > URL: https://issues.apache.org/jira/browse/HDDS-3167 > Project: Hadoop Distributed Data Store > Issue Type: Improvement >Reporter: Sammi Chen >Priority: Major > > For open container, it's replication is decided by Ratis pipeline number > size. For closed container, this is no such constrain. We can support more > replication factor cases, such as 2 factor, to save storage for some cold > data. -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: ozone-issues-unsubscr...@hadoop.apache.org For additional commands, e-mail: ozone-issues-h...@hadoop.apache.org
[jira] [Commented] (HDDS-3167) Support different replication factor for CLOSED Container
[ https://issues.apache.org/jira/browse/HDDS-3167?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17060778#comment-17060778 ] Marton Elek commented on HDDS-3167: --- > Marton Elek do you want to think it strictly and propose a more detail > solution . We can have more discussion here. What about writing a proposal together? I can write this part in more details, but I think your requirements and use cases are also important to be included. Just let me know which tool is better for you. Usually I use https://hackmd.io (plain markdown, easy to include) but fine with google docs or anything else... > Support different replication factor for CLOSED Container > - > > Key: HDDS-3167 > URL: https://issues.apache.org/jira/browse/HDDS-3167 > Project: Hadoop Distributed Data Store > Issue Type: Improvement >Reporter: Sammi Chen >Priority: Major > > For open container, it's replication is decided by Ratis pipeline number > size. For closed container, this is no such constrain. We can support more > replication factor cases, such as 2 factor, to save storage for some cold > data. -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: ozone-issues-unsubscr...@hadoop.apache.org For additional commands, e-mail: ozone-issues-h...@hadoop.apache.org
[jira] [Commented] (HDDS-3167) Support different replication factor for CLOSED Container
[ https://issues.apache.org/jira/browse/HDDS-3167?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17060041#comment-17060041 ] Sammi Chen commented on HDDS-3167: -- Hi [~elek] and [~swagle], thanks for the input. bq. How would you like to decide the replication factor for a specific container? Do you suggest to use one global settings or some kind of per-container settings? As Sid said, container is an internal abstraction, we cann't decide the replication factor of a specifc container from OM point of view. Also we might do this on SCM level, it would break the replication gurantee of OM keys. My initial idea is provide a archive replication factor on bucket level. once the container is moved from open to close, replication manager will ensure the archive replication factor is enforecd on each container. Also currently, bucket and container are not 1:1 binded. To simplify the logic, maybe we should allocate dedicated container for specific pipeline. But this will sacrifice write throughput. I like the storage class approach, it's more general and can handle wider scenarios. [~elek] do you want to think it strictly and propose a more detail solution :D . We can have more discussion here. > Support different replication factor for CLOSED Container > - > > Key: HDDS-3167 > URL: https://issues.apache.org/jira/browse/HDDS-3167 > Project: Hadoop Distributed Data Store > Issue Type: Improvement >Reporter: Sammi Chen >Priority: Major > > For open container, it's replication is decided by Ratis pipeline number > size. For closed container, this is no such constrain. We can support more > replication factor cases, such as 2 factor, to save storage for some cold > data. -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: ozone-issues-unsubscr...@hadoop.apache.org For additional commands, e-mail: ozone-issues-h...@hadoop.apache.org
[jira] [Commented] (HDDS-3167) Support different replication factor for CLOSED Container
[ https://issues.apache.org/jira/browse/HDDS-3167?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17058372#comment-17058372 ] Siddharth Wagle commented on HDDS-3167: --- Wouldn't such a setting make sense at a directory level if using FS semantics or bucket level for a pure object-store. Container sort of stands an internal abstraction, meaning how would an admin configure replication factor at container level? > Support different replication factor for CLOSED Container > - > > Key: HDDS-3167 > URL: https://issues.apache.org/jira/browse/HDDS-3167 > Project: Hadoop Distributed Data Store > Issue Type: Improvement >Reporter: Sammi Chen >Priority: Major > > For open container, it's replication is decided by Ratis pipeline number > size. For closed container, this is no such constrain. We can support more > replication factor cases, such as 2 factor, to save storage for some cold > data. -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: ozone-issues-unsubscr...@hadoop.apache.org For additional commands, e-mail: ozone-issues-h...@hadoop.apache.org
[jira] [Commented] (HDDS-3167) Support different replication factor for CLOSED Container
[ https://issues.apache.org/jira/browse/HDDS-3167?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17057818#comment-17057818 ] Marton Elek commented on HDDS-3167: --- How would you like to decide the replication factor for a specific container? Do you suggest to use one global settings or some kind of per-container settings? As a first step a generic settings seems to be enough, but what if somebody change the settings and restart the SCM. Should it be applied for all the remaining containers? If we would like to support EC in the future it would be great to support per-container replication settings. In fact, we need pairs of settings. For example: RATIS/THREE (when it's open) --> THREE (when it's closed) RATIS/ONE (when it's open) --> EC-5-2 (when it's closed) One approach to support this is introducing a the "storage class", similar to the S3 storage class. For example REDUCED = RATIS/ONE (for open) and THREE (for closed)) STANDARD = RATIS/THREE (open) and TWO (closed) INFREQUENT = RATIS/THREE (open) and EC (closed) It's not strictly about this issue, just thinking loudly here. > Support different replication factor for CLOSED Container > - > > Key: HDDS-3167 > URL: https://issues.apache.org/jira/browse/HDDS-3167 > Project: Hadoop Distributed Data Store > Issue Type: Improvement >Reporter: Sammi Chen >Priority: Major > > For open container, it's replication is decided by Ratis pipeline number > number. For closed container, this is no such constrain. We can support more > replication factor cases, such as 2 factor, to save storage for some cold > data. -- This message was sent by Atlassian Jira (v8.3.4#803005) - To unsubscribe, e-mail: ozone-issues-unsubscr...@hadoop.apache.org For additional commands, e-mail: ozone-issues-h...@hadoop.apache.org