Class CacheStats

java.lang.Object
org.frontcache.cache.CacheStats
public class CacheStats extends Object
A cheap size-and-churn snapshot of a cache processor, for the metrics exporter. Deliberately NOT getCacheStatus(), which the management API already returns: that one is a Map<String, String> built for human display, and reading it costs a fresh Lucene DirectoryReader every time. Fine for an operator clicking a console page; wrong for a Prometheus scrape arriving every few seconds forever. Everything here is read from state the process already holds.

Unknown is a value

Every field can be UNKNOWN, and the exporter registers no series for one that is. A cache processor with no on-disk tier does not have an "index size of 0" - it has no index, and a gauge reading a flat zero would be a lie an alert could be built on.

Micrometer-free on purpose

The cache package does not depend on the metrics package. Export is optional and off by default; having the cache implementations register meters directly would put an optional dependency on the request path's most central class. So the cache side exposes plain numbers and org.frontcache.metrics stays the only thing that knows what a meter is.

  • Field Summary

    Fields
    Modifier and Type
    Field
    Description
    static final CacheStats
    NOT_TRACKED
     
    static final long
    UNKNOWN
    The processor does not track this number.

  • Constructor Summary

    Constructors
    Constructor
    Description
    CacheStats(long l1Entries, long l1Evictions, long l2Entries, long l2DiskBytes)
    Kept for cache processors that do not track L1's byte size - CacheProcessor is a reflection-loaded extension point, so an implementation outside this repo still compiles and still reports honestly: not tracking a number is what UNKNOWN is for, and unlike the FallbackResolver case a default here cannot silently answer a question wrongly.
    CacheStats(long l1Entries, long l1Evictions, long l1ContentBytes, long l2Entries, long l2DiskBytes)
     

  • Method Summary

    Modifier and Type
    Method
    Description
    long
    getL1ContentBytes()
    Bytes of response CONTENT held in the in-memory tier, or UNKNOWN.
    long
    getL1Entries()
    Entries in the in-memory tier, or UNKNOWN.
    long
    getL1Evictions()
    Entries evicted from the in-memory tier since JVM start, or UNKNOWN.
    long
    getL2DiskBytes()
    Bytes the on-disk tier occupies, or UNKNOWN.
    long
    getL2Entries()
    Documents in the on-disk tier, or UNKNOWN.
    boolean
    isTracked(long value)
     
    String
    toString()
     

    Methods inherited from class Object

    clone, equals, finalize, getClass, hashCode, notify, notifyAll, wait, wait, wait

  • Field Details

    • UNKNOWN

      public static final long UNKNOWN
      The processor does not track this number. Distinct from zero.
      See Also:

    • NOT_TRACKED

      public static final CacheStats NOT_TRACKED

  • Constructor Details

    • CacheStats

      public CacheStats(long l1Entries, long l1Evictions, long l2Entries, long l2DiskBytes)
      Kept for cache processors that do not track L1's byte size - CacheProcessor is a reflection-loaded extension point, so an implementation outside this repo still compiles and still reports honestly: not tracking a number is what UNKNOWN is for, and unlike the FallbackResolver case a default here cannot silently answer a question wrongly.

    • CacheStats

      public CacheStats(long l1Entries, long l1Evictions, long l1ContentBytes, long l2Entries, long l2DiskBytes)

  • Method Details

    • getL1Entries

      public long getL1Entries()
      Entries in the in-memory tier, or UNKNOWN.

    • getL1Evictions

      public long getL1Evictions()
      Entries evicted from the in-memory tier since JVM start, or UNKNOWN. A monotonic total, not a rate: Prometheus derives the rate, and it is the rate that says whether L1 is too small for the working set.

    • getL1ContentBytes

      public long getL1ContentBytes()
      Bytes of response CONTENT held in the in-memory tier, or UNKNOWN. Not the tier's retained heap: header maps, key strings and per-object overhead are not in it. The distinction matters wherever this is shown next to a JVM heap figure. Maintained as the cache changes rather than measured on demand - see L1ByteCounter.

    • getL2Entries

      public long getL2Entries()
      Documents in the on-disk tier, or UNKNOWN.

    • getL2DiskBytes

      public long getL2DiskBytes()
      Bytes the on-disk tier occupies, or UNKNOWN. Answers "will the cache volume fill up", which the entry count does not - entry sizes on a page cache vary by orders of magnitude.

    • isTracked

      public boolean isTracked(long value)

    • toString

      public String toString()
      Overrides:
      toString in class Object