Re: [OpenAFS] Crash in volserver when restoring volume from backup.
Derrick Brashear wrote: looks like you already lost by the time it crashes. Program received signal SIGSEGV, Segmentation fault. [Switching to Thread 0x4aafd940 (LWP 30735)] 0x003c13078d60 in strlen () from /lib64/libc.so.6 (gdb) bt #0 0x003c13078d60 in strlen () from /lib64/libc.so.6 #1 0x00430092 in afs_vsnprintf (p=0x4aafc3ba 4BF+0, avail=999, fmt=value optimized out, ap=0x4aafc7c0) at ../util/snprintf.c:395 #2 0x00416a60 in vFSLog ( format=0x467838 1 Volser: ReadVnodes: IH_CREATE: %s - restore aborted\n, Just for the record, the original cause of this was due to a bug in TSM. IBM is currently working on it. -- Ragge ___ OpenAFS-info mailing list OpenAFS-info@openafs.org https://lists.openafs.org/mailman/listinfo/openafs-info
Re: [OpenAFS] GSoC Server priorities
-BEGIN PGP SIGNED MESSAGE- Hash: SHA1 Jake Thebault-Spieker wrote: On Thu, Oct 15, 2009 at 2:19 AM, Lars Schimmer l.schim...@cgv.tugraz.atwrote: -BEGIN PGP SIGNED MESSAGE- Hash: SHA1 Hi! I've seen the code for server priorities made in a GSoC is in the private windows test builds. As it could solve one smaller design problem for us, one question is open: On which base does it rate server near and far? AFAIK based on RTT time. Current private builds use RTT as taken from the rx statiscs that are gathered on a per-client basis, and use a log (base-e) scale to provide a rank based on the RTT. Our setup is expanded with a fileserver with only RO on a different subnet some miles away, attached to us via a static VPN. In one of our private subnets we drive our CAVE and this subnet is not routed accross the VPN - if we put the CAVE ROs on the fileserver on the far away fileserver, the clients does not reach them. For now, the ROs are not there (kinda bad in kind of backup reasons). Could this new feature solve this (in a bad way, I know) with just rate the bad fileserver horrible bad? If I understand what your end goals are, you're trying to minimize the amount of traffic to the file server on the other side of the VPN. If this is the case, then there is functionality available currently, using fs serverprefs. The administrator can set the server rank much higher than the ranks the other servers are being given, and this server will only be interacted with if all the other servers with lower ranks are down. The RTT based ranking only gets taken into account when there are rx statistics collected by the client, so if the client has never interacted with the file server on the other side of the VPN, there will be no rx statistics gathered, and the rank will default to the current scheme (based on which machine, subnet, network the server is on compared to the client). I hope this helps, please clarify if I misunderstood your goals. If you do try it, I would appreciate feedback you have. Thank you, that helped. Our point is: we got some clients in a 10.x subnet which is not routed across VPN ad therefore those clients cannot reach that fileserver. An as you tell, based on transmitted rx - with no traffic, no stats, no change for us. MfG, Lars Schimmer - -- - - TU Graz, Institut für ComputerGraphik WissensVisualisierung Tel: +43 316 873-5405 E-Mail: l.schim...@cgv.tugraz.at Fax: +43 316 873-5402 PGP-Key-ID: 0x4A9B1723 -BEGIN PGP SIGNATURE- Version: GnuPG v1.4.9 (GNU/Linux) Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org iEYEARECAAYFAkrYEfYACgkQmWhuE0qbFyMEgQCfRDJhizcbIMEmA+3kKuL6bT5P SdMAnjGbauLFSsvGR+kCRERbb5ILpVhO =WCsi -END PGP SIGNATURE- ___ OpenAFS-info mailing list OpenAFS-info@openafs.org https://lists.openafs.org/mailman/listinfo/openafs-info
Re: [OpenAFS] GSoC Server priorities
Lars Schimmer wrote: Thank you, that helped. Our point is: we got some clients in a 10.x subnet which is not routed across VPN ad therefore those clients cannot reach that fileserver. An as you tell, based on transmitted rx - with no traffic, no stats, no change for us. Lars: If you can explain what your goal is perhaps we can be more helpful. When the clients in the 10.x subnet ask the vlserver for the location of a volume, they will be told about all of the file servers that contain the desired volume. The clients will then attempt to probe the file servers to determine if the file servers are up or down. If a file server is unreachable, it will be marked down. Is your goal to permit the 10.x subnet to be routed across the VPN? Are you trying to determine if the 10.x clients will start to read data from the file server across the VPN if you did? Jeffrey Altman smime.p7s Description: S/MIME Cryptographic Signature
[OpenAFS] OpenAFS client memory usage
Hi, I have some questions about how OpenAFS 1.5.6400 works with RAM on Windows XP (SP2/SP3): While copying data from/to AFS, the afsd_service.exe increases it's working set and fills the cachefile (at least fs getcacheparms tells). On a client with a cache file of 384MB and 3GB RAM the mem usage in Task manager then grows to ~440MB and won't shrink until I either restart the service or start to run out of physical memory. I assume the VMM then tells the afsd_service to cleanup (correct?). While this isn't a problem on a client with 3GB of RAM it is indeed on older hosts with say 512MB of RAM. On such a client I experience paging when copying files. As far as I understood the afs clients works with the cache using a memory mapped file and opens views onto this file while using the cache. This views are then kept/cached in memory. Shouldn't the AFS client flush the contents to the disk periodically and close some of the views in order to release the memory again? Or is there a way to tell AFS to keep only e.g. 64MB of the cachefile in memory? Maybe we just missed something to configure. Or is there a recommendation for the cache file size in conjunction with amount of physical memory installed? At the moment we plan to workaround this effect by leaving the cache file size at the default value for all clients we distribute AFS to. Thanks for your hints, Nik ___ OpenAFS-info mailing list OpenAFS-info@openafs.org https://lists.openafs.org/mailman/listinfo/openafs-info
Re: [OpenAFS] OpenAFS client memory usage
Baumann Niklaus (LOG) wrote: Hi, I have some questions about how OpenAFS 1.5.6400 works with RAM on Windows XP (SP2/SP3): While copying data from/to AFS, the afsd_service.exe increases it's working set and fills the cachefile (at least fs getcacheparms tells). On a client with a cache file of 384MB and 3GB RAM the mem usage in Task manager then grows to ~440MB and won't shrink until I either restart the service or start to run out of physical memory. I assume the VMM then tells the afsd_service to cleanup (correct?). While this isn't a problem on a client with 3GB of RAM it is indeed on older hosts with say 512MB of RAM. On such a client I experience paging when copying files. As far as I understood the afs clients works with the cache using a memory mapped file and opens views onto this file while using the cache. This views are then kept/cached in memory. There is only one view. The cache is a paging file exactly the same as the Windows system paging file. Windows manages the memory utilization itself. The size of the paging space is determined by the size of the cache. Windows pulls into memory the portions of the paging file that are actively in use and flushes those that aren't. If you are writing to the cache, then there will be paging. This will be true on a machine with 3GB as well as a machine with 128MB. The cache uses a least recently used algorithm to select which objects (volumes, status, buffers, extents) to recycle. The objects that are going to be recycled are the ones that have no recently been used. That requires that Windows page in the memory for that region of the cache, accept the changes, and write the contents back to disk. Shouldn't the AFS client flush the contents to the disk periodically and close some of the views in order to release the memory again? Or is there a way to tell AFS to keep only e.g. 64MB of the cachefile in memory? This is dynamically managed by the Windows memory manager. Maybe we just missed something to configure. Or is there a recommendation for the cache file size in conjunction with amount of physical memory installed? At the moment we plan to workaround this effect by leaving the cache file size at the default value for all clients we distribute AFS to. Your cache size should be sufficiently large to store the working set of data. Otherwise, not only are you experiencing paging but you are going to be blocked on network I/O operations. I do not believe that you need to worry about the virtual memory size of the afsd_service.exe process. Jeffrey Altman smime.p7s Description: S/MIME Cryptographic Signature
[OpenAFS] Troubles with multihomed, nat'ed file server.
I've got a file server that is behind a nat and is also multihomed that is no longer working from outside the nat. It was working fine outside for a long while, though we're not sure when it stopped (It gets very light use externally). The server's NetInfo looks like this: 192.168.1.108 192.168.15.176 f 67.224.XX.XX A vos listaddr shows only the 192.168 addresses (and each is on its own line, though the docs make it sound like multihomed machines should have all address on the same line) On a whim I tried to nuke sysid and restart the fs process (thinking that by doing that it might update the vldb with all the addresses.) All this managed to do was break the fileserver with the following log messages: Fri Oct 16 12:10:40 2009 Creating new SysID file Fri Oct 16 12:10:40 2009 VL_RegisterAddrs rpc failed; The IP address exists on a different server; repair it I created a test volume on the server and manually did a vos changeloc to have it listed on the public interface. I can access that volume from the outside world just fine, but internal access is broken. vos listaddr still lists only the internal addresses. I know this is possible, what am I doing wrong? -Scott ___ OpenAFS-info mailing list OpenAFS-info@openafs.org https://lists.openafs.org/mailman/listinfo/openafs-info