×
Community Blog An In-Depth Understanding of Presto (2): Presto Memory Management

An In-Depth Understanding of Presto (2): Presto Memory Management

Part 2 of this series discusses Presto memory management.

By Yunlei
Contributed by Alibaba Cloud Storage

Part 1 of this series discussed Presto's architecture. Presto is a memory computing engine, which means it must implement fine-grained memory management to ensure the orderly and smooth execution of queries, avoiding cases such as starvation and deadlocks.

Memory Pool

Presto uses a logical memory pool to manage different types of memory requirements. Presto divides the entire memory into System Pool, Reserved Pool, and General Pool.

1

  1. System Pool is reserved for the system. 40% of memory space is reserved for the system by default.
  2. Reserved Pool and General Pool are used to allocate query runtime memory.
  3. General Pool is used for most queries. The largest query uses Reserved Pool, so the space of the Reserved Pool is equivalent to the maximum space used by a query to run on a machine. The default is the 10% space.
  4. General enjoys other memory spaces besides System Pool and General Pool.

Why Do We Use Memory Pools?

System Pool is used for the memory used by the system. For example, if data is passed between machines, the buffer is maintained in the memory. This part of the memory is mounted under the system name.

Why do we need to reserve memory? Why is the reserved memory equal to the maximum memory used by a query on the machine?

If there is no Reserved Pool, a query that consumes a lot of memory starts to run when the query is large and the memory space is almost occupied. However, there is no memory space for this query to run at this time, and this query has been mounted to wait for available memory. However, after other small memory queries finish running, new small memory queries are added. Small memory queries take up a small amount of memory, so it is easy to find available memory. As such, the large memory query hangs until it starves to death.

If you want to prevent the situation, a space must be reserved for the large memory query to run. The size of the reserved space is equal to the maximum memory that the query allows. Presto selects a query with the largest memory usage every second and allows it to use the reserved pool to avoid the situation where no available memory is there for the query to run.

Memory Management

2

Presto memory management is divided into two parts:

1.  Query Memory Management

  • A query is divided into many tasks. Each task has a thread loop to obtain the status of the task, including the memory used by the task. Aggregate the memory used for the query.
  • If the aggregate memory of a query exceeds a certain size, the query is forcibly terminated.

2.  Machine Memory Management

  • The coordinator has a thread that regularly rotates each machine and views the current machine memory status.

After the query and machine memory are aggregated, the coordinator selects the query with the largest memory usage and assigns it to the Reserved Pool.

Memory management is managed by the coordinator, which determines every second to specify that a query can use reserved memory on all machines. However, here is the problem. If the query is not run on a machine, is it a waste of memory reserved by the machine? Why not pick out the largest task execution on a single machine? The reason is deadlock. If the query has reserved memory on other machines, the execution will end soon. However, the task is not the largest on a machine and cannot be run all the time. As a result, the query fails to be ended.

Disclaimer: The views expressed herein are for reference only and don't necessarily represent the official views of Alibaba Cloud.

0 1 0
Share on

Alibaba Cloud Community

911 posts | 201 followers

You may also like

Comments