Resource allocation strategy for a use case

[Fred]

Hello everyone,

I would like to understand what action does the resource manager take for allocating the resource for the following scenario.

Lets say following are the parameters for a resource pool:
MEMEORYSIZE=MAXMEMORYSIZE=4G
Planned concurrency =4
Max concurrency=4

Lets say 4 queries are submitted simultaneously and as per these parameters each query will be allocated 1G for its use by the resource manager. Lets say, one of the queries query4 needs more than 1G memory for its execution. What will be the strategy of the resource manager in this case, if the other 3 queries are still running? Lets assume that the resource manager calculated the resource requirement for the query4, when the query was initially submitted, to be less than 1G so it does not go to the queue. Does the query get aborted in this case because of the resource unavailability, as I understand, queries once start running, will not go to queue again. Is my understanding correct or is there any other way this works?

Thanks, Fred

[NorbertKrupa]

My recommendation: test it out.

My hypothesis: depending on the timeout, it will get rejected as there are not enough resources available.

[scutter]

Hi Fred,

You are correct - once a query begins executing, it won’t queue. If the query needs additional memory but can’t get it (in this case because of the maxmemorysize on the resource pool), the query will immediately “fail”. But Vertica will also “retry” queries several times before they truly fail. If the additional memory was for a hash join, when the query retries it will execute with an alternative algorithm that requires less memory but is also typically slower. When the query retries, it may queue.

—Sharon

[Fred]

Thanks Sharon and Norbert for your inputs.

Fred