Modified BlockCacheManager to not create caches when size is zero.

[dlmarion]

No description provided.

[dlmarion]

I found this while working on #5302. I added a data block cache to the Compactor, but not index and summary caches. I noticed in the log that these caches were still being created.

[keith-turner]

The RFile client code has this NoopCache w/ a comment about leaking decompressors and it seems to use the NoopCache in lieu of null. I have no idea if this is relevant outside the client code. Depends on how things are closed in the server code overall, maybe the server code always cleans stuff up and closes decompressors and this is not a concern.

accumulo/core/src/main/java/org/apache/accumulo/core/client/rfile/RFileScanner.java

Lines 102 to 109 in bd9e676

	// This cache exist as a hack to avoid leaking decompressors. When the RFile code is not given a
	// cache it reads blocks directly from the decompressor. However if a user does not read all data
	// for a scan this can leave a BCFile block open and a decompressor allocated.
	//
	// By providing a cache to the RFile code it forces each block to be read into memory. When a
	// block is accessed the entire thing is read into memory immediately allocating and deallocating
	// a decompressor. If the user does not read all data, no decompressors are left allocated.
	private static class NoopCache implements BlockCache {

[dlmarion]

I noticed the NoopCache, but didn't think it would apply to the server side code. Your comment above made me look, and I'm not quite sure at this point. The code is pretty convoluted, but I think the comment in NoopCache still be correct on the client side. On the server side? no idea.

[dlmarion]

I also don't see where the existing BlockCache's close the Block such that BCFile.Reader.BlockReader.close is called.