Fix cached plan not getting invalidated

[DavIvek]

This PR is about resolving memory spike that happens when we use variable in MERGE statement. The MERGE clause that contains variable for a node will cause a significant memory spike 30x times.

[master < Task] PR

• Check, and update documentation if necessary
• Provide the full content or a guide for the final git message

To keep docs changelog up to date, one more thing to do:

• Write a release note here, including added/changed clauses
• Tag someone from docs team in the comments

closes #1251

[DavIvek]

If we avoid to use monotonic memory, the difference in memory consumption (when we use variable and when we don't) will decrease but still remain significant.

[gitbuda]

The problem here seems to be the caching of query plans (probably query plan cache). The solution is to implement some cache eviction strategy, LRU or some background process (LRU is the best place to start).

[DavIvek]

Since we are not invoking a garbage collector for plan cache, for testing purposes, I changed the Interpreter::PrepareIndexQuery part of the function where we remove cached plans from the skip list. Instead of calling remove for every cached plan, now we call the skip lists clear() function to force memory deallocation because remove() doesn't deallocate memory, it only marks skip list entry as deleted (and the garbage collector should do deallocation when it's safe to do it).

With that minor change, I was able to release all the memory used by the plan cache:
Firstly, by executing query: 'CREATE INDEX ON: MyNode1(address)', we will call plan_cache->clear(), following should remove all elements from the skip list.
After that, memory still isn't released, but if we execute the 'FREE MEMORY' query, memory usage should drop drastically.

I'm still confused about what actually happens there when we call the FREE MEMORY query.

[DavIvek]

TODO:

• rename query-plan-cache-ttl flag and think of what would be a good default value
• remove unnecessary parts of code

[antoniofilipovic]

Two comments left, and that is it

[antoniofilipovic]

Looks good to me, good job here 🥳