Re: switch to native locks by default?
Chris Hostetter wrote: : Ah, I hadn't realized that they might not be supported everywhere... I I'm just trusting the javadoc for NativeFSLockFactory ... i have no idea if it's accurate or not. Hi! I had added the caveat about native locks based on my dicey experience getting them working over NFS (NFS locking was not "turned on" by default in my setup; and, frustratingly, it would take ~35 seconds for a timeout to tell me this). I don't have any specific evidence that other OS/filesystems are problematic but then again I haven't done much research to understand overall portability of Java's native lock interface. It would not surprise me if other OS/filesystems had issues. I was hoping by getting the NativeFSLockFactory out there that it would get some healthy testing first and then we could use that feedback to decide whether benefits outweigh the risks of making it the default. It's not clear how many people have actually tested it at this point, though! : The current locking can also guard against mistakes though (multiple : instances of Solr trying to write to the same dir, someone opening a : Luke index on it, etc). right ... but it's only useful if all of the potential clients are using the same locking mechanism ... right now it's only safe to do any of those things if all the apps use SimpleFSLockFactory. all the more reason to make the factory and the lockDir configurable in Solr i guess. Mike
Re: switch to native locks by default?
: Ah, I hadn't realized that they might not be supported everywhere... I I'm just trusting the javadoc for NativeFSLockFactory ... i have no idea if it's accurate or not. : The current locking can also guard against mistakes though (multiple : instances of Solr trying to write to the same dir, someone opening a : Luke index on it, etc). right ... but it's only useful if all of the potential clients are using the same locking mechanism ... right now it's only safe to do any of those things if all the apps use SimpleFSLockFactory. all the more reason to make the factory and the lockDir configurable in Solr i guess. -Hoss
Re: switch to native locks by default?
On 1/14/07, Chris Hostetter <[EMAIL PROTECTED]> wrote: : Any reason we shouldn't switch to native locks by default? : In the future, read-only readers won't lock at all, but there will : still be a write lock for writers. just to clarify for people who may not be familiar (i had to go reread the LockFactory docs to remember) Native Locks may not be supported on all OSes, but have the perk of being cleaned up automatically in the event of a JVM crash. Ah, I hadn't realized that they might not be supported everywhere... I thought I saw some comments about switching Lucene to use that by default. Are you thinking we would add a new solrconfig.xml option to control this and default it to native locks? Not a bad idea... this is the first time i've noticed the SingleInstanceLockFactory ... in most cases couldn't we use that for Solr? I imagine we'll always need to support lock files if we wanted to make sure we can do some of hte "can i point Solr and an index created by _" use cases people have been asking about lately, ... but for the "normal" usage of Solr, isn't that good enough? The current locking can also guard against mistakes though (multiple instances of Solr trying to write to the same dir, someone opening a Luke index on it, etc). -Yonik
Re: switch to native locks by default?
: Any reason we shouldn't switch to native locks by default? : In the future, read-only readers won't lock at all, but there will : still be a write lock for writers. just to clarify for people who may not be familiar (i had to go reread the LockFactory docs to remember) Native Locks may not be supported on all OSes, but have the perk of being cleaned up automatically in the event of a JVM crash. Are you thinking we would add a new solrconfig.xml option to control this and default it to native locks? this is the first time i've noticed the SingleInstanceLockFactory ... in most cases couldn't we use that for Solr? I imagine we'll always need to support lock files if we wanted to make sure we can do some of hte "can i point Solr and an index created by _" use cases people have been asking about lately, ... but for the "normal" usage of Solr, isn't that good enough? -Hoss
switch to native locks by default?
Any reason we shouldn't switch to native locks by default? In the future, read-only readers won't lock at all, but there will still be a write lock for writers. -Yonik
