Hi Brett,

there have been changes to the internal scheduling of related work packages which may cause issues if you had cyclic dependencies in your data before 7.4. And as such, there have been quite a number of migrations from 7.3. to 7.4.

db/migrate/20120319135006_add_custom_field_translation_table.rb
db/migrate/20130807141542_remove_files_attached_to_projects_and_versions.rb
db/migrate/20131210113056_repair_invalid_default_work_package_custom_values.rb
db/migrate/20151028063433_boolearlize_bool_custom_values.rb
db/migrate/20170116105342_add_custom_options.rb
db/migrate/20170829095701_generate_wp_closure.rb
db/migrate/20171106074835_move_hashed_token_to_core.rb
db/migrate/20180105130053_rebuild_dag.rb
db/migrate/20180108132929_vacuum_relations.rb
db/migrate/20180116065518_add_hierarchy_paths.rb

If you can isolate these issues, we would very much appreciate a report to smoothen the ugprade path.

A downgrade is theoretically possible for those migrations, but this has not been done in production before for the relations rebuild as far as I know.

Best,
Oliver

Hi Oliver,

Thank you for the reply.

I ran a backup and then took a snapshot before I tried to rollback our installation. And unfortunately, like you pointed out, the SQL structure was broken and the rollback failed. So I reverted to my snapshot and we are limping along on 7.4.3 now.

We have a Debian Stretch VM running only OPM and another Debian VM running our MySQL/MariaDB database.

1. The performance since the upgrade is terrible. HTOP is showing 100% on 2 cores for SQL queries when work packages are clicked on/opened
2. My project manager has project templates he’s built and copies to new projects as we need them. Ever since the upgrade, when he does this, it makes exponential copies of copies and orders them all over the place

I know we should file bug reports and will soon. But I’m wondering if anyone else has experienced these issues too. Like I said, performance and relational copies worked great before the upgrade. And unfortunately we didn’t discover this until a couple weeks later so we’d lose a lot of progress if we tried to restore from a backup. We are actually considering forking and running 2 instances to eventually move away from 7.4.3

Another thing to note is we are running on VMWare on a Dell 710 poweredge server. Even though the resources we had before were fine, I doubled CPU and quadrupled RAM and it had no affect.

Thanks for any insight on this issue.

-Brett

Hi Brett,

thanks for the response and further insight. I’m assuming the migration was unable to resolve cyclic dependencies in your case and leaves 7.4. running in an unstable state. I’m confident we can resolve this and will reach out via email to get our team involved.

If you are interested in the technical details:

7.4. migrated away from simple relational because with the old structure still from redmine, it was possible to create cyclical dependencies that result in invalid to/from relations when involved with parent/child relationships since those do not share the same relational structure (the latter used nested sets, the former simple lookup tables).

With 7.4., we combine parent/child and all other relations into a closure table to represent the directed acyclic graph over all relation types to efficiently compute the transitive hull of the relations (and thus to decide whether a new relation will create a cyclic dependency).

This brings read optimization and comes at the cost of storage (we are storing a relation to each path in within a DAG subgraph) and having to find all cyclic dependencies in the migration and trying to resolve them. It is also the reason why we enforce PostgreSQL 9.5+ since we need UPSERTs as an optimization when (re-)writing relations

We obviously haven’t found all those cases in the migration and thus appreciate your feedback very much to approach and resolve it.

Best,
Oliver

Hi Bret,

is it possible for you to activate the slow queries log and when filing a bug report, add the logs to that report. This might give us the insights we need.

Regards,

Jens

Hello Oliver & Jens,

Thanks a lot for your interest in helping with this issue. We’ll definitely help you in anyway we can.

I’ll enable the ‘slow queries log’ and submit when filing the bug.

I’ll also lookout for that email as well.

Thanks again for your help.

Brett

Hi Brett (sorry for misspelling your name on my first reply),

please also take a look at an already existing bug and see if that also applies to you.

Regards

Jens