qng might suffer the OOM crash due to the memory leak of the go-ethereuem code cache.

[dindinw]

The issue is caused by ethereum/go-ethereum#26071

• The fastcache cause the memory leak, the go-ethereum might change to an new caching library that does not need an explicit close mechanism
  • the runtime.finalizer works for our OOM issue, but the current fix is not safe enough for us to use.
  • we have to replace the fastcache by ourself
  • or we might have to wait the issue to be resolved before we cherry-pick to our 1.21-qng
    • ex: trying to replace fastcache by the github.com/hashicorp/golang-lru/simplelru see core/state: replace fastcache code cache with gc-friendly structure ethereum/go-ethereum#26092
• The qng fully syncing might not be executed successfully.
• a temporary solution
  • make sure it has enough memory on the sync node.
    • 16G node is required for a fully syncing of the 1.25m blocks
    • big swap file should be created ( same with or bigger than the physical memory is recommended)
    • the virtual memory must be set to unlimited (ulimit -v unlimited), the fastcache use the unix mmap for the heap-off allocation.
  • watching the syncing process carefully
    • shutdown&restart the node if required. the normal shutdown is safe during the syncing. but OOM kill might result the data inconsistent.

[dindinw]

• core/state: replace fastcache code cache with gc-friendly structure ethereum/go-ethereum#26092 has been cherry-picked into go-ethereum-v1.10.21-qng branch
• the qng v1.0.18 release will include the bug fix with v1.10.21-qng.3 tag.
• the qng v1.0.19 dev branch will try to integrate the go-ethereum v1.11.x release.

[dindinw]

The cherry-picked PR26092 suffers the performance issue. we might wait for the future go-ethereum v1.11.0 release. it also means there is little possibility of picking all code changes into the old v.1.10.21 branch.

• common/lru: add generic LRU implementation ethereum/go-ethereum#26162
• common/lru: fix race in lru ethereum/go-ethereum#26164

[dindinw]

done at v1.0.18 release, see #300