Hi Tom, Good question. There is no airtight justification for not wiring the message, but I'll try anyway. First, it's OK to lose a message because recovery is attempted only for broken channels. If it wasn't for the successful attempt, all subsequent messages would be lost too. That explains why it's not wrong to return after every recovery attempt even successful ones. As for the why, ResilientOutputStream is the result of several iterations and the code evolved to its present state. While working on issue LBCORE-148, i.e. refactoring SocketAppender to use a resilient stream, returning after recovery attempts proved to be quite useful but that was just accidental, not by design. I hope this answers your question, On 13/04/2010 10:42 AM, Tom Liu wrote:
Hi logback-dev,
I wonder why ResilientOutputStreamBase does not try to write the log message after attempt the recovery? What's the consideration here?
public void write(byte b[], int off, int len) { if (isPresumedInError()) { if (!recoveryCoordinator.isTooSoon()) { attemptRecovery(); } return; // return regardless of the success of the recovery attempt }
try { os.write(b, off, len); postSuccessfulWrite(); } catch (IOException e) { postIOFailure(e); } }
Thanks very much, Tom