Ceph » CephFS

Can't fs top already change the sort order? I thought that was done in Neeraj's first tranche of improvements.

Greg Farnum wrote:

Can't fs top already change the sort order? I thought that was done in Neeraj's first tranche of improvements.

Another thing to notice here:
Before Neeraj's changes (first tranche of improvements), cephfs-top was displaying the clients in the descending order of resource usage. But that order doesn't exist anymore as it displays clients separately for each Filesystem now (say fs groups). That seems deviating from the original goal of having cephfs-top (highest resources usage on top).

As we are implementing top(1) like features for cephfs-top, top(1) always displays processes in the descending order of CPU usage. The 'f' menu in top(1) is just for selecting fields to display (Fields Management). As a result, it doesn't change the sort order in top(1) it's always descending order of CPU usage. So is it worth implementing for cephfs-top?

Venky Shankar wrote:

Jos Collin wrote:

Based on my discussion with Greg, I'm closing this ticket. Because the issue that the customer reported in BZ¹ is pagination and the tracker [2] is meant to resolve that issue. As per our discussion, the goal the customer has is to see all the clients and providing the limit option still restricts the number of clients which doesn't really solve the customer's issue.

[2] https://tracker.ceph.com/issues/55197

I disagree closing this tracker. The use of limit and order-by is to allow the user to list client according to a field (say descending order of cap hit ratio) and limit this by top N (say, 10) clients. Ins't that useful? With pagination, users would be able to scroll and view the entire client list, however, IMO, its useful to provide the user a way to sort and limit based on a particular field.

I think choosing the field based on which the sorting needs to be done will be taken care of by https://tracker.ceph.com/issues/55463 and desc order can be kept as default.But the order-by can handle if someone wants it to be in another order.

Jos Collin wrote:

Greg Farnum wrote:

Can't fs top already change the sort order? I thought that was done in Neeraj's first tranche of improvements.

Another thing to notice here:
Before Neeraj's changes (first tranche of improvements), cephfs-top was displaying the clients in the descending order of resource usage. But that order doesn't exist anymore as it displays clients separately for each Filesystem now (say fs groups). That seems deviating from the original goal of having cephfs-top (highest resources usage on top).

As we are implementing top(1) like features for cephfs-top, top(1) always displays processes in the descending order of CPU usage. The 'f' menu in top(1) is just for selecting fields to display (Fields Management). As a result, it doesn't change the sort order in top(1) it's always descending. So is it worth implementing for cephfs-top?

I don't think cephfs-top was displaying the clients in the descending order of resource usage.Venky mentioned the client connection order,see the description of https://tracker.ceph.com/issues/55463

Neeraj Pratap Singh wrote:

Jos Collin wrote:

Greg Farnum wrote:

Can't fs top already change the sort order? I thought that was done in Neeraj's first tranche of improvements.

Another thing to notice here:
Before Neeraj's changes (first tranche of improvements), cephfs-top was displaying the clients in the descending order of resource usage. But that order doesn't exist anymore as it displays clients separately for each Filesystem now (say fs groups). That seems deviating from the original goal of having cephfs-top (highest resources usage on top).

As we are implementing top(1) like features for cephfs-top, top(1) always displays processes in the descending order of CPU usage. The 'f' menu in top(1) is just for selecting fields to display (Fields Management). As a result, it doesn't change the sort order in top(1) it's always descending. So is it worth implementing for cephfs-top?

I don't think cephfs-top was displaying the clients in the descending order of resource usage.Venky mentioned the client connection order,see the description of https://tracker.ceph.com/issues/55463

There is no point in having a cephfs-top tool, if it doesn't display the top clients. As per top(1) it should be a constant field say wsp(MB/s), IMO. Then if Fields Management is there, then the user can add/remove fields. Still the Fields Management doesn't make sense for lesser number of fields that cephfs-top has (only 20 fields, which the screen accommodates mostly).

Having a `sort-by-field` option is handy for the point I mentioned in https://tracker.ceph.com/issues/55121#note-4. The `limit` functionality is kind of closely tied with `sort-by` (using `limit` without `sort-by` does not really makes sense).

Jos Collin wrote:

Greg Farnum wrote:

Can't fs top already change the sort order? I thought that was done in Neeraj's first tranche of improvements.

Another thing to notice here:
Before Neeraj's changes (first tranche of improvements), cephfs-top was displaying the clients in the descending order of resource usage. But that order doesn't exist anymore as it displays clients separately for each Filesystem now (say fs groups). That seems deviating from the original goal of having cephfs-top (highest resources usage on top).

As we are implementing top(1) like features for cephfs-top, top(1) always displays processes in the descending order of CPU usage. The 'f' menu in top(1) is just for selecting fields to display (Fields Management). As a result, it doesn't change the sort order in top(1) it's always descending order of CPU usage. So is it worth implementing for cephfs-top?

My understanding was wrong here. There is an 's' command to reset the default sort field.