Response time when querying assets is proportional to the number of roles a user possesses


Issue: In our production system our LTI users have upwards of 30 roles, resulting in response times of at least 15 seconds whenever an event's metadata/scheduling information is updated or conflict checks are performed.

For non-privileged users, roles determine whether a user has read/write access to an asset. This determination takes place within the AssetManagerWithSecurity class, via a database/persistence query which considers the underlying asset and all of the (requesting) user's roles.

Currently, all the roles possessed by a user are used to construct the query above; this results in the query blow-out described by MH-12641.

Attached is a snapshot of a New Relic event summary of the latest conflict check performed by an end-user, where the response time is 39.8s.

Ideally, Asset Manager should filter out those roles not relevant to the requested asset before performing the DB query.


duncan smith
March 1, 2018, 1:31 PM

It would be great if roles not relevant to the mediapackage are filtered out before commencing with the DB query.

I was thinking something along the lines of

duncan smith
March 1, 2018, 1:02 PM

A single conflict check by an end-user today took 92s to complete. Query below:

duncan smith
February 28, 2018, 12:22 PM

The codepath relevant to the query's construction can be found in on lines 179 thru 193.

Excuse the format below, but I can't seem to setup formatting on this server:




duncan smith