fix: read cache in page manager drop newest page

[HarroYuz]

What does this PR do?

Fix a bug. When the size of the readCache in PageManager exceeds its limit, it is supposed to remove the oldest pages, as indicated by the comments. However, when comparing the removal priority between pages, the page with a larger lastAccessed timestamp (newer one) is given higher priority.

Checklist

• I have run the build using mvn clean package command
• My unit tests cover both failure and success scenarios

[lvca]

Thanks for your first PR!

Checking the condition, it still seems correct to me: the sorted tree is ordered by getLastAccessed() + getPhysicalSize(). So the 1st element in the tree is the page with the smallest long as last accessed = older.

Instead, now that I see the algo, the page size should be inverted: with the same last accessed timestamp, it should remove the page with the largest size. This is a very rare case that shouldn't impact performance in the real world, but it should be inverted.

[HarroYuz]

Thank you for your response.

I agree with you on the removal order that the page with smaller last accessed and larger physical size should be removed first.

However, from the code segment below the declearation of oldestPages, I found that the 1st page in the tree is regarded as the newest one while the last page is regarded as the oldest one:

PageManager.java:401-418

for (final CachedPage page : readCache.values()) {
      if (oldestPagesRAM < ramToFree) {
        // FILL FIRST PAGES
        oldestPages.add(page);
        oldestPagesRAM += page.getPhysicalSize();
      } else {
        if (page.getLastAccessed() < oldestPages.last().getLastAccessed()) {
          oldestPages.add(page);
          oldestPagesRAM += page.getPhysicalSize();

          // REMOVE THE LESS OLD
          final Iterator<CachedPage> it = oldestPages.iterator();
          final CachedPage pageToRemove = it.next();
          oldestPagesRAM -= pageToRemove.getPhysicalSize();
          it.remove();
        }
      }
    }

In this case, the 1st element in the tree should be the page with largest lastAccessed and smaller physicalSize.

[lvca]

I think that piece of code is completely non-sense :-) If the pages are ordered, they should just be purged in sequence until the threshold is reached. Working on it.

[lvca]

Closing this PR and opening an new issue: #1287.