kestrel 源码分析之三 PersistentQueue对应一个内存中的队列

PersistentQueue类主要用于在内存中维护一个队列，该类是Kestrel的核心类，是真正对于一个队列的维护。我们看到上文中QueueCollection中主要是调用下面的setup方法，下面我们看看setup方法到底做了什么

def setup() {
    synchronized {
      queueSize = 0
      replayJournal()
    }
  }

def replayJournal() {
    if (!config.keepJournal) return

    log.info("Replaying transaction journal for '%s'", name)
    xidCounter = 0

    journal.replay {
      case JournalItem.Add(item) =>
        _add(item)
        // when processing the journal, this has to happen after:
        if (!journal.inReadBehind && queueSize >= config.maxMemorySize.inBytes) {
          log.info("Dropping to read-behind for queue '%s' (%d bytes)", name, queueSize)
          journal.startReadBehind()
        }
      case JournalItem.Remove => _remove(false, None)
      case JournalItem.RemoveTentative(xid) =>
        _remove(true, Some(xid))
        xidCounter = xid
      case JournalItem.SavedXid(xid) => xidCounter = xid
      case JournalItem.Unremove(xid) => _unremove(xid)
      case JournalItem.ConfirmRemove(xid) => openTransactions.remove(xid)
      case JournalItem.Continue(item, xid) =>
        openTransactions.remove(xid)
        _add(item)
      case x => log.error("Unexpected item in journal: %s", x)
    }

    log.info("Finished transaction journal for '%s' (%d items, %d bytes) xid=%d", name, queueLength,
             journal.size, xidCounter)
    journal.open()

    // now, any unfinished transactions must be backed out.
    for (xid <- openTransactionIds) {
      journal.unremove(xid)
      _unremove(xid)
    }
  }

我们看到setup方法主要是回放journal文件中的redo Log。我们看到该类主要是读Journal文件，并解析为JournalItem对象，并根据对象类型对PersistentQueue进行入队和出队。我们看到一共有7种不同的日志类型，下面我们看看几个主要的类型日志的结构
ADDX：
Remove：

我们会看到在回放日志文件的同时，他会记录所有到目前为止所有的没有结束的事务在openTransactions中，其中他们的key的集合为openTransactionIds，对这些没有结束的事务做unremove操作。
另外还有一个关键是在对队列进行Add和Remove操作时调用checkRotateJournal（）方法，即检查是否需要对日志文件进行切分、删除等操作。下面我们看看具体的代码

// you are holding the lock, and config.keepJournal is true.
  private def checkRotateJournal() {
    /*
     * if the queue is empty, and the journal is larger than defaultJournalSize, rebuild it.
     * if the queue is smaller than maxMemorySize, and the combined journals are larger than
     *   maxJournalSize, rebuild them. (we are not in read-behind.)
     * if the current journal is larger than maxMemorySize, rotate to a new file. if the combined
     *   journals are larger than maxJournalSize, checkpoint in preparation for rebuilding the
     *   older files in the background.
     */
    if ((journal.size >= config.defaultJournalSize.inBytes && queueLength == 0) ||
        (journal.size + journal.archivedSize > config.maxJournalSize.inBytes &&
         queueSize < config.maxMemorySize.inBytes)) {
      log.info("Rewriting journal file for '%s' (qsize=%d)", name, queueSize)
      journal.rewrite(openTransactionIds.map { openTransactions(_) }, queue)
    } else if (journal.size > config.maxMemorySize.inBytes) {
      log.info("Rotating journal file for '%s' (qsize=%d)", name, queueSize)
      val setCheckpoint = (journal.size + journal.archivedSize > config.maxJournalSize.inBytes)
      journal.rotate(openTransactionIds.map { openTransactions(_) }, setCheckpoint)
    }
  }

1. 当内存中queue为空，并且日志文件的大小大于defaultJournalSize时，将创建一个新的日志文件并删除该时间点之前的文件。
2. 当之前的日志文件大小加上当前日志文件大小大于maxJournalSize，并且queue中数据的大小小于maxMemorySize时，将创建一个新的日志文件并删除该时间点之前的文件。
3. 当日志文件的大小大于maxMemorySize，将创建新的日志文件，之后达到的消息将存入新的文件。