DVLS Cache

avatar

We seem to have a cahing problem when using a load balancer with two dvls node. Before i go futher i must say that we did not configure afinity as the the documentation says that the solution is "stateless" (Deploy in a high-availability or load-balancing environment - Devolutions Documentation).

Let say i create a new vault and the load balancer redirect me to the first node for the creation of the vault. The vault is created and i can see it in the database. Now i want to update the vault. If i am lucky i will be directed to the first node but if not i won't be able to modify the vault as it will appear but not show any information when i hit modify. The easyest way to reproduce the problem:

Passing trought the LB you create a vault. Then refresh the page until you fall on the second node where the new vault won't appear. Hit refresh again and the vault may appear if you fall on the first node. After a couple of minute, the vault will always appear for both node.

You can also open two session, one on each node without the LB. Create the vault on one node and then on the other session/node consult the list of vault and the new vault won't show (even pressing the refresh button AND refreshing the page).

Is there a way to resolve this behavior apart configuring afinity? Let say we configure the afinity it will work until a failover occur. And when the failover occur we will fall with the same problem for all objects created in the last ?? minutes (i don't have the exact time it takes for the second node to fetch the new information)

Regards,
Pierre Ayotte

All Comments (4)

avatar

Hello Pierre,

Thank you for reaching out to us regarding this,

I appreciate your detailed message, could you confirm which version of DVLS you are currently using?

I'm also wondering if you started encountering this issue recently or if this has been occurring since you configured the Load balancing.

Let us know,

Best regards,

Samuel Dery

avatar

Version is 2023.3.13.0 and the situation only appear with the load balancer

avatar

Hi Pierre,

I am sorry for the confusion, you should use server affinity. As mentioned the server relies heavily on server caching and there might be delays for some data to be fetched from the database again. As an exemple for the vaults it may take up to five minutes before the data is synced.

I will suggest a correction in the documentation on this subject.

Best regards,

Benoît Sansregret

avatar

Hello Benoit, thanks i agree that it should be clearer in the documentation. Now did configured the afinity in the F5 and it is working good. For those who used F5 i found that configuring the pool advance option "Action On Service Down" to "Reject" so when a node goes down it will fall over other node rapidly and seemlessly.