Roxy 1.7.5 Issue

[rskainth82]

In order to deploy 8.0-6.4 version of MarkLogic, we are required to upgrade the Roxy version that we have been using at our organization Aetna. The most recent version of Roxy is 1.7.5, but its giving us issues, when we tested it with current ML version:8.0-5.6 that we have.

New version of Roxy 1.7.5 which is required for the MarkLogic upgrade (8.0-6.4) has a new feature,
#616 Replica forests: Suboptimal initial layout + incorrect layout on Rebootstrap after scaleout

As part of this feature, Roxy code is updated to create the replica forest name automatically without considering the assignment provided in the configuration file. Also the feature is released to update the replica forests distribution across the cluster deprecating the old round robin fashion. The host in which replica forests need to be created is decided by a formula automatically in the Roxy overriding the configuration provided in the config file.

This feature creates additional replica forests in the hosts with a new name. This is going to affect the original configuration.

We want to check whether there is an option to disregard this automatic calculation by Roxy for replica forests as we want Roxy to use assignment configuration from the configuration file, instead of doing this on its own. As otherwise it will change the way replicas have been setup in our environments. We don't want to change our current setup unless its really required to.

Which Operating System are you using?
Red Had Linux 6.8 (Santiago)
Which version of MarkLogic are you using?
8.0-5.6
Which version of Roxy are you using (see version.txt)?
1.7.5

We know that at the minimum, we will be required to go to Roxy 1.7.4 in order for us to upgrade to MarkLogic 8.0-6.4, but since Roxy 1.7.5 is already out and it addresses some of the issues that were present in Roxy 1.7.4, we wanted to use Roxy 1.7.5, but this new version is giving us issues with creation of replica forests. We are currently looking into using 1.7.4, but since 1.7.5 was already out and that is why we were looking into using this as it has some bug fixes from 1.7.4.
If possible, we would need to the bug fix or enhancement for this issue as soon as possible, so that we can use the latest and greatest version of Roxy in our applications.

[RobertSzkutak]

Have you considered using the -apply-changes option when you bootstrap to avoid touching forest configuration altogether?

Your concerns are valid. I will do a little digging and see what I can do to put your mind at ease.

[rskainth82]

Hi Rob, —apply-changes option won’t work for us, since we have our deployment process automated and —apply-changes needs WHAT changes we need to apply. Since our process is automated, we can’t change it and supply a different value, every time we have some changes to apply. Please let us know, if you found something else, which can help us or let us know when will this issue be fixed.

[RobertSzkutak]

My suggestion was that you could use --apply-changes to bootstrap everything except forest configuration. In other words, bootstrap would work like normal except that it would just skip the forest configuration piece entirely. You wouldn't need to change the value of --apply-changes in your process every time.

Hope that clears up my suggestion. I'm continuing a technical investigation into this but I can't commit to any timeline of when that will conclude. If you're working with any ML consultants or SEs, I recommend having them look into this on your behalf to expedite finding a solution.

[grtjn]

Bit late to this discussion, but wanted to point out that the automatic layout is only applied to databases which have a <forests-per-host> element in ml-config.xml. By default that applies to content-db, but you can append that to other databases, or perhaps more useful here: remove that tag from content-db, and do the full forest mapping yourself with ml-config.xml..

[grtjn]

I'd be interested to learn, though, if there are differences in the forest layout between latest and older versions. It should not be the case that mere upgrade of Roxy causes a lot of forest changes. Ideally none (except when there are hosts added or removed and such)..

[RobertSzkutak]

There's quite a bit of detail along with discussion of this issue in the PR which made the changes being referenced here: #689

CC @jamsilvia

[grtjn]

Thanks, still need to catch up on that long PR..

[rskainth82]

just an update, we are not using forests-per-host tag anywhere in ml-config.xml. All the forest assignments are under assignments tag and then these forests are just referred using the IDs under the different databases configurations.

[rskainth82]

@RobertSzkutak We can't use your suggestion of using --apply-changes to bootstrap everything except forest configuration, as we wan't to be able to bootstrap forest level changes as well using the configuration files. We don't want to be manually doing and maintaining these changes.

We will wait for a more robust technical solution to our problem. Till then, we are going to use Roxy 1.7.4 version instead of Roxy 1.7.5