Calling `issueFullFileCheck()` does not seem to remove old unreferenced objects

[andriikovalov]

Environment Details

• MicroStream Version: 08.01.01-MS-GA
• JDK version: 17
• OS: linux

Describe the bug

When I reset and store the application root, after calling issueFullFileCheck() the old inaccessible data seems to be still in the storage (judging from the size).

To Reproduce

public class App {
    public static void main(String[] args) throws IOException {
        Path storagePath = Files.createTempDirectory("microstream");
        final EmbeddedStorageManager storageManager = EmbeddedStorage.start(storagePath);

        for (int i = 0; i < 50; i++) {
            storageManager.setRoot(new byte[1000000]); // 1 MB
            storageManager.storeRoot();
        }

        storageManager.issueFullFileCheck();
        storageManager.shutdown();

        System.out.println("Storage size " + getSize(storagePath));  // Expected ~1 MB, actual ~50 MB
    }

    public static long getSize(Path dir) throws IOException {
        return Files.walk(dir).map(Path::toFile).filter(File::isFile).mapToLong(File::length).sum();
    }
}

Expected behavior

The storage is shrinked to only contain the current root.

Additional context

I observe the same behaviour when I wrap my byte array into a root object, and repeatedly re-initialize the array and call storeRoot().

class Root {
    public byte[] data;
}

[hg-ms]

Hello,
This not a bug. In your example the storage has no time to clean up old data.
The point of time when data gets deleted depends on several factors, the most important ones are:

• The Java GC
• The storage GC
• The storage GC configuration
• The storage internal cache
• The write load.
  In your example you create a heavy write load that requires some tweaking of the storage housekeeping.

The Example below should perform better regarding the cleanup.
It ensures that the Java GC and storage GC are executed and sets a very small object live time for the storage cache and increases the time budget for housekeeping.

final EmbeddedStorageManager storageManager = EmbeddedStorage
	.start(Storage.ConfigurationBuilder()
		.setEntityCacheEvaluator(Storage.EntityCacheEvaluator(1000, 10))
		.setHousekeepingController(Storage.HousekeepingController(100, 1000_000_000))
		.setStorageFileProvider(Storage.FileProvider(storagePath)).createConfiguration());

for (int i = 0; i < 50; i++) {
	storageManager.setRoot(new byte[1000000]); // 1 MB
	storageManager.storeRoot();
}

System.gc();
storageManager.issueFullGarbageCollection();
storageManager.issueFullFileCheck();

[andriikovalov]

Okay, thank you. I thought that explicitly calling housekeeping would delete unreachable data (as mentioned in #179), but in fact it is not guaranteed, and cannot be enforced, right?
Your snippet with "aggressive housekeeping" also gives me the same result (storage not shrinked).