Dear whiskers. According to the merging of these two PRs [1][2], not the memory consumption for the prewarm pool will be included in user memory.
Previously when you have a machine with 16GB memory, you had to consider the memory consumption of the prewarm pool. For example, if you assign 4GB of memory for your prewarm pool, you can't configure more than 10GB of memory as userMemory. (2GB are reserved for the invoker itself). If you configure more memory to userMemory, it might cause OOM due to the sum of userMemory and memory for the prewarm pool can exceed the limit. Now, you can just configure the userMemory as 14GB and assign any size of memory to the prewarm pool as long as it does not exceed 14GB. In case there are too many containers, the system would not create a new prewarm container to fill the pool. Instead, the shortage would be backfilled by a periodic adjustment. In short, you may want to increase your userMemory config for the existing OW deployment to include the memory for prewarm pool. -dom [1] https://github.com/apache/openwhisk/pull/4911 [2] https://github.com/apache/openwhisk/pull/5048
