From: Wolfgang M. <wol...@gm...> - 2005-11-21 20:58:24
|
Just a quick summary on our findings concerning the read-only issue: I had the suspicion that the file locks we introduced with the last snapshot are not always released properly on Linux. I tested a fresh installation on three different machines, but could not observe any locking issues. The file locks were always correctly released when the jvm shut down. Joseph also reports that he might have started a second server without checking if the previously started process was still running. This would explain why the database switched to read-only mode. I'm thus no longer sure if we really have a bug here. If someone experiences similar issues again, please report them. And please keep in mind: if the database does not allow you to create a collection or does otherwise tell you that it is read-only, make sure that no other eXist process is still running in the background. Older versions of eXist allowed a second process to open the same database files. This was harmless in most cases. However, with the introduction of the logging & recovery code, conflicting writes to the journal will render a completely unusable database. That's the reason why we have to write-lock all files. If a write-lock is detected during startup, the database will either switch to read-only mode (Linux) or terminate with an error (Windows). On Linux, the log files will contain a message: "database runs in read-only mode". Wolfgang On 11/21/05, Wolfgang Meier <wol...@gm...> wrote: > Hi, > > > Is there something about the port number of the url "/exist" that the > > database actually needs, or is this an irrelevant problem? > > No, the port should not be relevant. I also received reports about > read-only databases from other users, and they all seem to point > towards locks not being released. I will try to fix this issue later > today, though I'm not yet sure how Java's file locking works on Linux > and what might be wrong with it. > > Wolfgang > |