Functions that traverse the IProfilingBlockInputStream tree may have performance issues when the tree contains SharedQueryBlockInputStream

[windtalker]

Bug Report

The same issue as #4494
The root cause is when IProfilingBlockInputStream tree contains SharedQueryBlockInputStream, the tree is actually a DAG, and when traverse the DAG as a tree, it need much redundant traverse, at worst case, the time complexity become O(m^n), where n is the number of SharedQueryBlockInputStream in the DAG, m is the number of children for the tree node.
Currently, the functions that may have this problem inlcudes:

• IProfilingBlockInputStream::setProcessListElement
• IProfilingBlockInputStream::setProgressCallback

[SeaRise]

It looks like collectCPUTimeNsImpl won't have this problem because cpu_time_ns_collected is used in collectCPUTimeNs to avoid calling collectCPUTimeNsImpl repeatedly

[windtalker]

OK, I've updated the issue description.

[yibin87]

Can we fix the issue by override these interfaces in SharedQueryBlockInputStream, and check if redundant ops?

[SeaRise]

Can we fix the issue by override these interfaces in SharedQueryBlockInputStream, and check if redundant ops?

Because these two functions are useless in TiFlash, they can be omitted.