I have managed to be in the Dynamics CRM/365 world for over 7 years without having to write a single plugin on Retrieve and Retrieve Multiple. The recommendation that I give is to stay away from those. The reason is simple, it sounds horrible from a performance standpoint, and even people from Microsoft have recommended against it in many scenarios. Faced with an issue recently where we had to really consider it, I did some research and testing to try to measure the impact of such plugins on system performance. This article provides some background as to why we recommend against these types of plugins, and it also provides some of our finding after we tested for performance.
Why are Plugins on Retrieve Multiple scary?
When looking at the event execution pipeline for Dynamics CRM/365, we need to consider that there are a lot of steps involved as part of every CRM transactions. To do anything, we need to go through the CRM web service APIs which will start the chain of events in the pipeline (pre-validate, pre-event, core action/database access and then post-event).
This means that in general, it is a good practice to build everything for optimized performance to give your user base a good experience. It’s not like having a custom database where you can create store procedures easily, add triggers and so on, taking advantage of the SQL Server features and infrastructure.
Now, back to Retrieve and Retrieve Multiple plugins.
Retrieve Multiple: Think about it this way, you retrieve a list of 200 accounts, your plugin on Retrieve Multiple fires once and gives you the list of accounts being returned to the screen with all columns being retrieved. For each of these row, you do some type of operation. It doesn’t too sound good, does it?
Retrieve: You double click on an account from a list view. As the account columns are being retrieved to display the account form on the screen, you plugin on Retrieve fires and gives you the account object. At that point you can modify the content of the columns being retured as required before they are returned to the screen for the user. This really doesn’t sound too bad.
Some Findings on the impact on Performance
To provide some context into what we were trying to do, I wrote about multi-language lookup in a previous blog post. One of the solution that we have considered for one of our client is to use a plugin on Retrieve and Retrieve Multiple in order to change the value of the lookup primary fields in order to display a value in the user’s current language. The method we used is similar to what Aileen Gusni does here and almost identical to what Scott Durow does here.
We store a Region in a custom Entity. Accounts have a lookup that indicates its region.
The Region’s Primary Field contains a concatenation of the English and French region names with a relatively safe separator (we use “|”). When you load the list view of accounts, we have a plugin on Retrieve Multiple that looks at the columns being returned. If the Region column is returned, we retrieve the user’s language, we split the name of the region with the separator (the name is available in the entity reference) and we replace its value in the target object by the region name in the user’s language. The plugin on Retrieve does the same operation on the single role being retrieved.
The average execution time for the Retrieve Multiple plugin when loading 150 rows was 15.43 milliseconds so 0.01543 seconds.
The average execution time for the Retrieve plugin to load one rows was too little for the system to return a value (we got 0 milliseconds every time).
The Region entity has English Name and French Name attributes. When you load the list view of accounts, we have a plugin on retrieve multiple that looks at the columns being returned. If the Region column is returned, we retrieve the user’s language, we then retrieve the English or French name from the Region entity and we replace its value in the target object by the region name in the user’s language. The plugin on Retrieve does the same operation on the single role being retrieved.
The average execution time for the Retrieve Multiple plugin to load 150 rows was 1003.94 milliseconds so 1.00394 seconds.
The average execution time for the Retrieve plugin to load one rows was 16.02 milliseconds so 0.01602 seconds.
What should you read into this?
While these numbers don’t look too crazy at all, especially in the first scenario, there are a lot of factors to take into consideration that are not really showing here and that will vary in almost any scenario.
- What is the infrastructure you are running on? The faster your severs and networks, the better the performance will be.
- What you do in these plugins matters a great deal. You should avoid or limit the number of read/writes to the database during the execution of those plugins.
- Our tests were made with a low level of users in the system, it is critical to scale up and see what these numbers look like at peak time of your system.
With all this said, I still recommend against it. Use with a great deal of caution! If/when possible, use calculated fields instead of writing plugins on these messages. This will also keep you away from limitation such as this one.
Hope this helps!