Re: Cyrus crashed on redundant platform - need better availability?
Hi, Ken Murchison wrote: I wouldn't hold out hope of anything being available in some months. I wrote my replication code two years ago, and submitted it to Rob and Ken about this time last year. Neither I or they have put any significant work into the code since then. As I indicated in my previous message, we all have other priorities right now. I can imagine, but I hoped that priorities would change a bit with the amount of users that repeatedly This link appears dead. All I get is To clipboard. Oops. There was never supposted to be a link :-) interest in this feature and with the money we are willing to put in :-| I'm willing to work on it if there is money available. You are the only one that has says that you would commit money. Where are the rest of the folks? Based on the number of people that stepped up to pay for virtdomains support (zero), I'm guessing there are fewer out there willing to spend money than you think. But I could be wrong. I'm happy to see that there are indeed others interested in this ;-) Other than the altnamespace project ($5000) that I did for a (unnamed) company in Texas, Jeremy Howard at Fastmail is the only one who has consistently paid for features. I'll let him disclose what he has spent, if he chooses to, but its safe to say that its been more than just pizza and beer. I expected more then pizza and beer, so that's no surprise :-) I'd have to look at David's patch again and discuss things with CMU to get a good time estimate, but I'm guessing that a project like this would cost a few thousand dollars. Ok, I'll start a discussion with our management based on your latest estimation ($3000-$5000) and I'll let you know about the results. (Might take a while, I think at least not this week. If you have more details (for instance time estimation) let me know.) Bye, Paul --- Cyrus Home Page: http://asg.web.cmu.edu/cyrus Cyrus Wiki/FAQ: http://cyruswiki.andrew.cmu.edu List Archives/Info: http://asg.web.cmu.edu/cyrus/mailing-list.html
Re: Cyrus crashed on redundant platform - need better availability?
Hi, --On Freitag, 10. September 2004 16:27 Uhr +0200 Paul Dekkers [EMAIL PROTECTED] wrote: Right, works fine for us for the most part. Hasn't always been like that, but the most recent kernel updates by Red Hat have improved matters a lot. What did the kernel improve? memory management for the most part. With 8 GB of RAM and lots of it free there were previously situations where either the cache grew too large, causing the machine to become extremely slow, or where forks failed (even though there were oodles of free RAM). Both seem to have been resolved in 2.4.9-e.49enterprise. You are not using a clustered filesystem, right? No. Although many on the list claim that this (having 2 boxes with 1 disk-array) is a nice way for redundancy I'm in doubt now if this is true. It's good but not perfect. We recently installed a huge SAN and are now in the process of moving over the mail data to reside there. Fibrechannel seems to be much more error tolerant than SCSI. Hmm, I don't expect the problems to be SCSI-related. Maybe it has to do with GEOM and SMP in FreeBSD 5.2.1, but not the SCSI-bus itself. (There are two seperate controllers for both machines, they never see each other on the same SCSI bus...) That's not what I was talking about. We have a similar setup, yet still there were instances when Red Hat's cluster software failed to write to the shared storage. I guess this was caused by the slow-downs connected to the memory management, but Red Hat support indicated that shared storage connected via FibreChannel would not have been as susceptible to these problems. --On Freitag, 10. September 2004 21:36 Uhr +0200 Jure PeÄ?ar [EMAIL PROTECTED] wrote: The kernel that shipped with RedHat AS 2.1 was useless for most of the tasks i tried it with. About three revisions later it became somewhat more usefull for non-oracle types of use, but i've rolled my own and am not following the state of it now. That's fine if you don't have to rely on commercial support. Our management decided to go the supported path all the way. That doesn't leave you many options. I have to say that when it works, the cluster software works extremely well. It's just that it hasn't always worked in the past ... ;-) I haven't had problems with the fiber itself, i've only had lots of fun with the firmware on the disks themselves and some with the qlogic drivers. We've had our share of problems with those as well, but I hear that Red Hat AS 3.0 ships with working QLogic drivers that work out of the box. Cheers, Sebastian Hagedorn -- Sebastian Hagedorn M.A. - RZKR-R1 (Gebäude 52), Zimmer 18 Zentrum für angewandte Informatik - Universitätsweiter Service RRZK Universität zu Köln / Cologne University - Tel. +49-221-478-5587 pgpmuwLb5sS5G.pgp Description: PGP signature
Re: Cyrus crashed on redundant platform - need better availability?
Hi, Sebastian Hagedorn wrote: You are not using a clustered filesystem, right? No. I can imagine that would be one of the advantages of RH's clustering, since you don't have to mount a filesystem in that case for a machine that just crashed - it would safe time... But I suppose RH's cluster manager takes care of mounting the partitions and checking them if there are any errors. It's good but not perfect. We recently installed a huge SAN and are now in the process of moving over the mail data to reside there. Fibrechannel seems to be much more error tolerant than SCSI. Where you working with a multi-initiator enviroment (as RH calls it) or single initiator (e.g. with 2 machines on exactly the same SCSI bus, or two seperate interfaces on your array's SCSI controller?) I think with a multi-initiator enviroment (as we have it) there is a very limited chance of failures. Hmm, I don't expect the problems to be SCSI-related. Maybe it has to do... That's not what I was talking about. We have a similar setup, yet still there were instances when Red Hat's cluster software failed to write to the shared storage. I guess this was caused by the slow-downs connected to the memory management, but Red Hat support indicated that shared storage connected via FibreChannel would not have been as susceptible to these problems. Do you think using RH's cluster software is a valuable consideration for this kind of clustering setup? Using FreeBSD there are not that many clustering solutions for now, and if it's advisable to at least consider using RH here (although I have no experience with RH) we can certainly look at it. (Any idea how fast RH would recover services?) On the other hand, if there is a application level redundancy on its way, it doesn't really matter on what platform the machine runs, so it would still make me happier and even with FreeBSD. And I would rather put my money there. Even if it means we'll have to wait for some months, we would do that and take the risk of running on a less automatic-failover-situation with a worst-case downtime of 30 mins (or 2 mins regulary with sync-mounted filesystems now). The kernel that shipped with RedHat AS 2.1 was useless for most of the tasks i tried it with. About three revisions later it became somewhat more usefull for non-oracle types of use, but i've rolled my own and am not following the state of it now. That's fine if you don't have to rely on commercial support. Our management decided to go the supported path all the way. That doesn't leave you many options. I have to say that when it works, the cluster software works extremely well. It's just that it hasn't always worked in the past ... ;-) That's a plus for RH (ES|AS) 3 Paul --- Cyrus Home Page: http://asg.web.cmu.edu/cyrus Cyrus Wiki/FAQ: http://cyruswiki.andrew.cmu.edu List Archives/Info: http://asg.web.cmu.edu/cyrus/mailing-list.html
Re: Cyrus crashed on redundant platform - need better availability?
On Wed, 15 Sep 2004, Paul Dekkers wrote: On the other hand, if there is a application level redundancy on its way, it doesn't really matter on what platform the machine runs, so it would still make me happier and even with FreeBSD. And I would rather put my money there. Even if it means we'll have to wait for some months, I wouldn't hold out hope of anything being available in some months. I wrote my replication code two years ago, and submitted it to Rob and Ken about this time last year. Neither I or they have put any significant work into the code since then. As I indicated in my previous message, we all have other priorities right now. -- David Carter Email: [EMAIL PROTECTED] University Computing Service,Phone: (01223) 334502 New Museums Site, Pembroke Street, Fax: (01223) 334679 Cambridge UK. CB2 3QH. --- Cyrus Home Page: http://asg.web.cmu.edu/cyrus Cyrus Wiki/FAQ: http://cyruswiki.andrew.cmu.edu List Archives/Info: http://asg.web.cmu.edu/cyrus/mailing-list.html
Re: Cyrus crashed on redundant platform - need better availability?
David Carter wrote: On Wed, 15 Sep 2004, Paul Dekkers wrote: On the other hand, if there is a application level redundancy on its way, it doesn't really matter on what platform the machine runs, so it would still make me happier and even with FreeBSD. And I would rather put my money there. Even if it means we'll have to wait for some months, I wouldn't hold out hope of anything being available in some months. I wrote my replication code two years ago, and submitted it to Rob and Ken about this time last year. Neither I or they have put any significant work into the code since then. As I indicated in my previous message, we all have other priorities right now. I can imagine, but I hoped that priorities would change a bit with the amount of users that repeatedly http://www.interglot.com/toclipboard.php?b=1d=2t=herhaaldelijks=herhaaldelijkw=repeatedlyshowed interest in this feature and with the money we are willing to put in :-| Paul --- Cyrus Home Page: http://asg.web.cmu.edu/cyrus Cyrus Wiki/FAQ: http://cyruswiki.andrew.cmu.edu List Archives/Info: http://asg.web.cmu.edu/cyrus/mailing-list.html
Re: Cyrus crashed on redundant platform - need better availability?
On Wed, 15 Sep 2004 13:38:43 +0200 Paul Dekkers [EMAIL PROTECTED] wrote: But I suppose RH's cluster manager takes care of mounting the partitions and checking them if there are any errors. Not really, at least not by itself. See http://people.redhat.com/jrfuller/cms/ for detailed documentation of what is included with RH AS 2.1 (it's some $500 extra for AS 3). I had to write some pretty paranoid scripts that take care of assembling software raids, checking the fs and mountig it while taking care about the other machine to prevent problems. Of course all this would be much easier with some kind of clustered fs, but clustered fs brings a new problem: locking. Almost all i've seen so far have an external 'locking manager' on a separate box, which brings ethernet latency into every lock operation, which i'm sure is very noticable in the lock-heavy usage patterns as mail is. But this is just my feeling, i haven't yet benchmarked any :) Do you think using RH's cluster software is a valuable consideration for this kind of clustering setup? Using FreeBSD there are not that many clustering solutions for now, and if it's advisable to at least consider using RH here (although I have no experience with RH) we can certainly look at it. (Any idea how fast RH would recover services?) This RH cluster software is nothing fancy; i'm sure equivalents exists for BSDs. See documentation link above. Actually it is just Kimberlite (http://oss.missioncriticallinux.com/projects/kimberlite/), sold with RedHat support. Speed of recovery is almost completely out of the cluster control. The only thing that matters for the cluster is what your cyrus init script returns when called with 'status' parameter. Everything else is up to your init scripts. Of course, if one box dies completely, the other takes over in the configurable time. -- Jure Pear --- Cyrus Home Page: http://asg.web.cmu.edu/cyrus Cyrus Wiki/FAQ: http://cyruswiki.andrew.cmu.edu List Archives/Info: http://asg.web.cmu.edu/cyrus/mailing-list.html
Re: Cyrus crashed on redundant platform - need better availability?
Hi, --On Mittwoch, 15. September 2004 13:38 Uhr +0200 Paul Dekkers [EMAIL PROTECTED] wrote: You are not using a clustered filesystem, right? No. I can imagine that would be one of the advantages of RH's clustering, since you don't have to mount a filesystem in that case for a machine that just crashed - it would safe time... I'm not sure if Red Hat even supports a clustered FS at this time. It certainly didn't when we set up the system more than two years ago. But I suppose RH's cluster manager takes care of mounting the partitions and checking them if there are any errors. Right. The unmounting/mounting of partitions usually works fine, but there have been problems at times. The worst one was causing alternating crashes of both nodes: sd(8,73)): ext3_free_blocks: Freeing blocks not in datazone - block = 225139276, count = 1 EXT3-fs error (device sd(8,73)): ext3_free_blocks: Freeing blocks not in datazone - block = 1919637002, count = 1 EXT3-fs error (device sd(8,73)): ext3_free_blocks: Freeing blocks not in datazone - block = 894788200, count = 1 EXT3-fs error (device sd(8,73)): ext3_free_blocks: Freeing blocks not in datazone - block = 1883792719, count = 1 EXT3-fs error (device sd(8,73)): ext3_free_blocks: Freeing blocks not in datazone - block = 1347113037, count = 1 EXT3-fs error (device sd(8,73)): ext3_free_blocks: Freeing blocks not in datazone - block = 829312330, count = 1 EXT3-fs error (device sd(8,73)): ext3_free_blocks: Freeing blocks not in datazone - block = 893538370, count = 1 EXT3-fs error (device sd(8,73)): ext3_free_blocks: Freeing blocks not in datazone - block = 1450341715, count = 1 EXT3-fs error (device sd(8,73)): ext3_free_blocks: Freeing blocks not in datazone - block = 909390198, count = 1 EXT3-fs error (device sd(8,73)): ext3_free_blocks: Freeing blocks not in datazone - block = 1366706293, count = 1 EXT3-fs error (device sd(8,73)): ext3_free_blocks: Freeing blocks not in datazone - block = 846548333, count = 1 EXT3-fs error (device sd(8,73)): ext3_free_blocks: Freeing blocks not in datazone - block = 1630746450, count = 1 EXT3-fs error (device sd(8,73)): ext3_free_blocks: Freeing blocks not in datazone - block = 860649837, count = 1 EXT3-fs error (device sd(8,73) leading to this: Assertion failure in journal_forget_Rsmp_094dfde7() at transaction.c:1226: !jh-b_committed_data [ cut here ] kernel BUG at transaction.c:1226! invalid operand: Kernel 2.4.9-e.38enterprise CPU:3 EIP:0010:[f885b636]Not tainted EFLAGS: 00010282 EIP is at journal_forget_Rsmp_094dfde7 [jbd] 0xd6 eax: 0025 ebx: ce6e8c10 ecx: c02f7f84 edx: 0008dad9 esi: cd95f3e0 edi: cd7a3094 ebp: cd7a3000 esp: cb947d70 ds: 0018 es: 0018 ss: 0018 Process ctl_cyrusdb (pid: 4500, stackpage=cb947000) Stack: f8863f30 04ca e7b08b20 cd95f3e0 cd7a3000 000b cd95f3e0 f885ee69 ce14ac40 cd95f3e0 cd95f3e0 cd95f3e0 cab35900 ce14ac40 f886bc8c ce14ac40 0002 cd95f3e0 cd95f3e0 cd6ad000 cd6ae000 cdd93000 cd95f3e0 0002 Call Trace: [f8863f30] .LC7 [jbd] 0x0 (0xcb947d70) [f885ee69] journal_revoke_Rsmp_56fa5ece [jbd] 0xf9 (0xcb947d8c) [f886bc8c] ext3_forget [ext3] 0x7c (0xcb947da8) [f886df3a] ext3_free_branches [ext3] 0xda (0xcb947dd8) [f886df2c] ext3_free_branches [ext3] 0xcc (0xcb947e30) [f886e2ec] ext3_truncate [ext3] 0x2bc (0xcb947e74) [f885a285] start_this_handle [jbd] 0x125 (0xcb947eac) [f885a38f] journal_start_Rsmp_ec53be73 [jbd] 0xbf (0xcb947ec4) [f886bd5e] start_transaction [ext3] 0x4e (0xcb947ee4) [f886bee7] ext3_delete_inode [ext3] 0xe7 (0xcb947f08) [f887a080] ext3_sops [ext3] 0x0 (0xcb947f28) [c015dd1c] iput_free [kernel] 0x14c (0xcb947f2c) [f886f9c3] ext3_lookup [ext3] 0x73 (0xcb947f40) [c015addb] dentry_iput [kernel] 0x4b (0xcb947f50) [c01541ab] vfs_unlink [kernel] 0x1eb (0xcb947f60) [c0152c41] lookup_hash [kernel] 0x91 (0xcb947f6c) [c015427a] sys_unlink [kernel] 0x9a (0xcb947f88) [c01181c0] do_page_fault [kernel] 0x0 (0xcb947fb0) [c01073e3] system_call [kernel] 0x33 (0xcb947fc0) Code: 0f 0b 59 58 53 e8 40 03 00 00 8b 43 24 c7 43 14 00 00 00 00 0Kernel panic: not continuing I had to intercept the boot process manually before the cluster software starts and fsck the partition. Not good. But this problem has been fixed in a kernel update. It's good but not perfect. We recently installed a huge SAN and are now in the process of moving over the mail data to reside there. Fibrechannel seems to be much more error tolerant than SCSI. Where you working with a multi-initiator enviroment (as RH calls it) or single initiator (e.g. with 2 machines on exactly the same SCSI bus, or two seperate interfaces on your array's SCSI controller?) I think with a multi-initiator enviroment (as we have it) there is a very limited chance of failures. I'm not sure about the terminology, but we have two separate SCSI busses on the RAID, one for each host. I thought that was single initiator? The problem that regularly occurred is the
Re: Cyrus crashed on redundant platform - need better availability?
Hi, --On Mittwoch, 15. September 2004 13:38 Uhr +0200 Paul Dekkers [EMAIL PROTECTED] wrote: You are not using a clustered filesystem, right? No. I can imagine that would be one of the advantages of RH's clustering, since you don't have to mount a filesystem in that case for a machine that just crashed - it would safe time... I'm not sure if Red Hat even supports a clustered FS at this time. It certainly didn't when we set up the system more than two years ago. I thinks that's exactly why they bought Sistina with GFS - and GPL'd it. Does anybody know how it works with cyrus-imapd? Simon --- Cyrus Home Page: http://asg.web.cmu.edu/cyrus Cyrus Wiki/FAQ: http://cyruswiki.andrew.cmu.edu List Archives/Info: http://asg.web.cmu.edu/cyrus/mailing-list.html
Re: Cyrus crashed on redundant platform - need better availability?
Paul Dekkers wrote: David Carter wrote: On Wed, 15 Sep 2004, Paul Dekkers wrote: On the other hand, if there is a application level redundancy on its way, it doesn't really matter on what platform the machine runs, so it would still make me happier and even with FreeBSD. And I would rather put my money there. Even if it means we'll have to wait for some months, I wouldn't hold out hope of anything being available in some months. I wrote my replication code two years ago, and submitted it to Rob and Ken about this time last year. Neither I or they have put any significant work into the code since then. As I indicated in my previous message, we all have other priorities right now. I can imagine, but I hoped that priorities would change a bit with the amount of users that repeatedly http://www.interglot.com/toclipboard.php?b=1d=2t=herhaaldelijks=herhaaldelijkw=repeatedlyshowed This link appears dead. All I get is To clipboard. interest in this feature and with the money we are willing to put in :-| I'm willing to work on it if there is money available. You are the only one that has says that you would commit money. Where are the rest of the folks? Based on the number of people that stepped up to pay for virtdomains support (zero), I'm guessing there are fewer out there willing to spend money than you think. But I could be wrong. Other than the altnamespace project ($5000) that I did for a (unnamed) company in Texas, Jeremy Howard at Fastmail is the only one who has consistently paid for features. I'll let him disclose what he has spent, if he chooses to, but its safe to say that its been more than just pizza and beer. I'd have to look at David's patch again and discuss things with CMU to get a good time estimate, but I'm guessing that a project like this would cost a few thousand dollars. -- Kenneth Murchison Oceana Matrix Ltd. Software Engineer 21 Princeton Place 716-662-8973 x26 Orchard Park, NY 14127 --PGP Public Key--http://www.oceana.com/~ken/ksm.pgp --- Cyrus Home Page: http://asg.web.cmu.edu/cyrus Cyrus Wiki/FAQ: http://cyruswiki.andrew.cmu.edu List Archives/Info: http://asg.web.cmu.edu/cyrus/mailing-list.html
Re: Cyrus crashed on redundant platform - need better availability?
also take a look at the heartbeat package at linux-ha.org This works on linux, *BSD, and solaris (there were people working on a AIX port, but they apparently dropped it shortly before finishing) David Lang On Wed, 15 Sep 2004, Jure [UTF-8] PeÄ~Mar wrote: Date: Wed, 15 Sep 2004 17:07:20 +0200 From: Jure [UTF-8] PeÄ~Mar [EMAIL PROTECTED] To: Paul Dekkers [EMAIL PROTECTED] Cc: [EMAIL PROTECTED], [EMAIL PROTECTED] Subject: Re: Cyrus crashed on redundant platform - need better availability? On Wed, 15 Sep 2004 13:38:43 +0200 Paul Dekkers [EMAIL PROTECTED] wrote: But I suppose RH's cluster manager takes care of mounting the partitions and checking them if there are any errors. Not really, at least not by itself. See http://people.redhat.com/jrfuller/cms/ for detailed documentation of what is included with RH AS 2.1 (it's some $500 extra for AS 3). I had to write some pretty paranoid scripts that take care of assembling software raids, checking the fs and mountig it while taking care about the other machine to prevent problems. Of course all this would be much easier with some kind of clustered fs, but clustered fs brings a new problem: locking. Almost all i've seen so far have an external 'locking manager' on a separate box, which brings ethernet latency into every lock operation, which i'm sure is very noticable in the lock-heavy usage patterns as mail is. But this is just my feeling, i haven't yet benchmarked any :) Do you think using RH's cluster software is a valuable consideration for this kind of clustering setup? Using FreeBSD there are not that many clustering solutions for now, and if it's advisable to at least consider using RH here (although I have no experience with RH) we can certainly look at it. (Any idea how fast RH would recover services?) This RH cluster software is nothing fancy; i'm sure equivalents exists for BSDs. See documentation link above. Actually it is just Kimberlite (http://oss.missioncriticallinux.com/projects/kimberlite/), sold with RedHat support. Speed of recovery is almost completely out of the cluster control. The only thing that matters for the cluster is what your cyrus init script returns when called with 'status' parameter. Everything else is up to your init scripts. Of course, if one box dies completely, the other takes over in the configurable time. -- Jure Peÿÿar --- Cyrus Home Page: http://asg.web.cmu.edu/cyrus Cyrus Wiki/FAQ: http://cyruswiki.andrew.cmu.edu List Archives/Info: http://asg.web.cmu.edu/cyrus/mailing-list.html -- There are two ways of constructing a software design. One way is to make it so simple that there are obviously no deficiencies. And the other way is to make it so complicated that there are no obvious deficiencies. -- C.A.R. Hoare --- Cyrus Home Page: http://asg.web.cmu.edu/cyrus Cyrus Wiki/FAQ: http://cyruswiki.andrew.cmu.edu List Archives/Info: http://asg.web.cmu.edu/cyrus/mailing-list.html
Re: Cyrus crashed on redundant platform - need better availability?
how much are you asking for? David Lang On Wed, 15 Sep 2004, Ken Murchison wrote: Date: Wed, 15 Sep 2004 11:44:45 -0400 From: Ken Murchison [EMAIL PROTECTED] To: Paul Dekkers [EMAIL PROTECTED] Cc: David Carter [EMAIL PROTECTED], [EMAIL PROTECTED] Subject: Re: Cyrus crashed on redundant platform - need better availability? Paul Dekkers wrote: David Carter wrote: On Wed, 15 Sep 2004, Paul Dekkers wrote: On the other hand, if there is a application level redundancy on its way, it doesn't really matter on what platform the machine runs, so it would still make me happier and even with FreeBSD. And I would rather put my money there. Even if it means we'll have to wait for some months, I wouldn't hold out hope of anything being available in some months. I wrote my replication code two years ago, and submitted it to Rob and Ken about this time last year. Neither I or they have put any significant work into the code since then. As I indicated in my previous message, we all have other priorities right now. I can imagine, but I hoped that priorities would change a bit with the amount of users that repeatedly http://www.interglot.com/toclipboard.php?b=1d=2t=herhaaldelijks=herhaaldelijkw=repeatedlyshowed This link appears dead. All I get is To clipboard. interest in this feature and with the money we are willing to put in :-| I'm willing to work on it if there is money available. You are the only one that has says that you would commit money. Where are the rest of the folks? Based on the number of people that stepped up to pay for virtdomains support (zero), I'm guessing there are fewer out there willing to spend money than you think. But I could be wrong. Other than the altnamespace project ($5000) that I did for a (unnamed) company in Texas, Jeremy Howard at Fastmail is the only one who has consistently paid for features. I'll let him disclose what he has spent, if he chooses to, but its safe to say that its been more than just pizza and beer. I'd have to look at David's patch again and discuss things with CMU to get a good time estimate, but I'm guessing that a project like this would cost a few thousand dollars. -- Kenneth Murchison Oceana Matrix Ltd. Software Engineer 21 Princeton Place 716-662-8973 x26 Orchard Park, NY 14127 --PGP Public Key--http://www.oceana.com/~ken/ksm.pgp --- Cyrus Home Page: http://asg.web.cmu.edu/cyrus Cyrus Wiki/FAQ: http://cyruswiki.andrew.cmu.edu List Archives/Info: http://asg.web.cmu.edu/cyrus/mailing-list.html -- There are two ways of constructing a software design. One way is to make it so simple that there are obviously no deficiencies. And the other way is to make it so complicated that there are no obvious deficiencies. -- C.A.R. Hoare --- Cyrus Home Page: http://asg.web.cmu.edu/cyrus Cyrus Wiki/FAQ: http://cyruswiki.andrew.cmu.edu List Archives/Info: http://asg.web.cmu.edu/cyrus/mailing-list.html
Re: Cyrus crashed on redundant platform - need better availability?
Paul Dekkers wrote: David Carter wrote: On Wed, 15 Sep 2004, Paul Dekkers wrote: On the other hand, if there is a application level redundancy on its way, it doesn't really matter on what platform the machine runs, so it would still make me happier and even with FreeBSD. And I would rather put my money there. Even if it means we'll have to wait for some months, I wouldn't hold out hope of anything being available in some months. I wrote my replication code two years ago, and submitted it to Rob and Ken about this time last year. Neither I or they have put any significant work into the code since then. As I indicated in my previous message, we all have other priorities right now. I can imagine, but I hoped that priorities would change a bit with the amount of users that repeatedly http://www.interglot.com/toclipboard.php?b=1d=2t=herhaaldelijks=herhaaldelijkw=repeatedlyshowed This link appears dead. All I get is To clipboard. interest in this feature and with the money we are willing to put in :-| I'm willing to work on it if there is money available. You are the only one that has says that you would commit money. Where are the rest of the folks? Based on the number of people that stepped up to pay for virtdomains support (zero), I'm guessing there are fewer out there willing to spend money than you think. But I could be wrong. Other than the altnamespace project ($5000) that I did for a (unnamed) company in Texas, Jeremy Howard at Fastmail is the only one who has consistently paid for features. I'll let him disclose what he has spent, if he chooses to, but its safe to say that its been more than just pizza and beer. I'd have to look at David's patch again and discuss things with CMU to get a good time estimate, but I'm guessing that a project like this would cost a few thousand dollars. We are very interested in replicated shared folders. We have different cyrus-imapd servers in different countries and would like to have common shared folders. If this could also be implemented I'm sure we were able to help sponsoring it. There are also a number of commercial vendors of cyrus-imapd based solutions who should be very interested in application level replication. Simon --- Cyrus Home Page: http://asg.web.cmu.edu/cyrus Cyrus Wiki/FAQ: http://cyruswiki.andrew.cmu.edu List Archives/Info: http://asg.web.cmu.edu/cyrus/mailing-list.html
Re: Cyrus crashed on redundant platform - need better availability?
Simon Matter wrote: Hi, --On Mittwoch, 15. September 2004 13:38 Uhr +0200 Paul Dekkers [EMAIL PROTECTED] wrote: You are not using a clustered filesystem, right? No. I can imagine that would be one of the advantages of RH's clustering, since you don't have to mount a filesystem in that case for a machine that just crashed - it would safe time... I'm not sure if Red Hat even supports a clustered FS at this time. It certainly didn't when we set up the system more than two years ago. I thinks that's exactly why they bought Sistina with GFS - and GPL'd it. Does anybody know how it works with cyrus-imapd? If you are interested in using a shared filesystem on a SAN for server redundancy, then you could try using a replicated Murder (Cyrus 2.3). Such a config is running at a local University using 4 Sun servers and QFS (Sun's SAN filesystem) on a Hitachi fibre array. I haven't tested this with GFS, but if it has correct file locking and memory mapping support, then it might work. I'm fairly confident that SGI's XFS would work, although I haven't tried it. -- Kenneth Murchison Oceana Matrix Ltd. Software Engineer 21 Princeton Place 716-662-8973 x26 Orchard Park, NY 14127 --PGP Public Key--http://www.oceana.com/~ken/ksm.pgp --- Cyrus Home Page: http://asg.web.cmu.edu/cyrus Cyrus Wiki/FAQ: http://cyruswiki.andrew.cmu.edu List Archives/Info: http://asg.web.cmu.edu/cyrus/mailing-list.html
Re: Cyrus crashed on redundant platform - need better availability?
David Lang wrote: how much are you asking for? Since this is probably as complex, if not more, as altnamespace, I'd say somewhere between $3000-$5000 as an initial estimate. That's 30-50 hours at a fairly cheap rate. If people want to start pledging their support, perhaps enough incentive can be pooled. If people don't feel comfortable doing this in public, then feel free to send me a private email. On Wed, 15 Sep 2004, Ken Murchison wrote: Date: Wed, 15 Sep 2004 11:44:45 -0400 From: Ken Murchison [EMAIL PROTECTED] To: Paul Dekkers [EMAIL PROTECTED] Cc: David Carter [EMAIL PROTECTED], [EMAIL PROTECTED] Subject: Re: Cyrus crashed on redundant platform - need better availability? Paul Dekkers wrote: David Carter wrote: On Wed, 15 Sep 2004, Paul Dekkers wrote: On the other hand, if there is a application level redundancy on its way, it doesn't really matter on what platform the machine runs, so it would still make me happier and even with FreeBSD. And I would rather put my money there. Even if it means we'll have to wait for some months, I wouldn't hold out hope of anything being available in some months. I wrote my replication code two years ago, and submitted it to Rob and Ken about this time last year. Neither I or they have put any significant work into the code since then. As I indicated in my previous message, we all have other priorities right now. I can imagine, but I hoped that priorities would change a bit with the amount of users that repeatedly http://www.interglot.com/toclipboard.php?b=1d=2t=herhaaldelijks=herhaaldelijkw=repeatedlyshowed This link appears dead. All I get is To clipboard. interest in this feature and with the money we are willing to put in :-| I'm willing to work on it if there is money available. You are the only one that has says that you would commit money. Where are the rest of the folks? Based on the number of people that stepped up to pay for virtdomains support (zero), I'm guessing there are fewer out there willing to spend money than you think. But I could be wrong. Other than the altnamespace project ($5000) that I did for a (unnamed) company in Texas, Jeremy Howard at Fastmail is the only one who has consistently paid for features. I'll let him disclose what he has spent, if he chooses to, but its safe to say that its been more than just pizza and beer. I'd have to look at David's patch again and discuss things with CMU to get a good time estimate, but I'm guessing that a project like this would cost a few thousand dollars. -- Kenneth Murchison Oceana Matrix Ltd. Software Engineer 21 Princeton Place 716-662-8973 x26 Orchard Park, NY 14127 --PGP Public Key--http://www.oceana.com/~ken/ksm.pgp --- Cyrus Home Page: http://asg.web.cmu.edu/cyrus Cyrus Wiki/FAQ: http://cyruswiki.andrew.cmu.edu List Archives/Info: http://asg.web.cmu.edu/cyrus/mailing-list.html -- Kenneth Murchison Oceana Matrix Ltd. Software Engineer 21 Princeton Place 716-662-8973 x26 Orchard Park, NY 14127 --PGP Public Key--http://www.oceana.com/~ken/ksm.pgp --- Cyrus Home Page: http://asg.web.cmu.edu/cyrus Cyrus Wiki/FAQ: http://cyruswiki.andrew.cmu.edu List Archives/Info: http://asg.web.cmu.edu/cyrus/mailing-list.html
Re: Cyrus crashed on redundant platform - need better availability?
On Wed, Sep 15, 2004 at 02:07:08PM -0400, Ken Murchison wrote: David Lang wrote: how much are you asking for? Since this is probably as complex, if not more, as altnamespace, I'd say somewhere between $3000-$5000 as an initial estimate. That's 30-50 hours at a fairly cheap rate. If people want to start pledging their support, perhaps enough incentive can be pooled. If people don't feel comfortable doing this in public, then feel free to send me a private email. I'm certainly interested in adding some redundancy to our Cyrus installation. I'm about to upgrade the hardware to a single Sun V480 with 4 1200 MHz CPUs and 16 gigs of memory. The two internal disks will be mirrored, and contain only the OS files. Everything else will be on external RAID arrays. The next expansion should add more IMAP storage and provide redundancy in the case of software or equipment failure. I'm aware of Murder, but I'm not sure that it's the best solution for us. I don't control the funding, but I can recommend something. -- -Gary Mills--Unix Support--U of M Academic Computing and Networking- --- Cyrus Home Page: http://asg.web.cmu.edu/cyrus Cyrus Wiki/FAQ: http://cyruswiki.andrew.cmu.edu List Archives/Info: http://asg.web.cmu.edu/cyrus/mailing-list.html
Paying for developers? (was: Re: Cyrus crashed on redundant platform - need better availability?)
Jure Pear wrote: I still think that it would be best to have two filesystems instead of one, so with mirroring on application level (cyrus)... :-) I'd rather see murder store a message on two sepparate machines ... Actually to have duplicated mailboxes in sync over a pool of backend machines, with murder taking care of backlogs when one of them would go down. So many users cried for this feature (to provide not just horizontal scalability with murder, but to have redundant backends which can hold each others replicas too) that I wonder: if it's so important to us, the cyrus users, why don't we collect some money and pass it to the developers? Maybe it could help to make the implementation real, and the developers have already demonstrated that they can design and code such things. -- Attila Nagy e-mail: [EMAIL PROTECTED] Free Software Network (FSN.HU) phone @work: +361 371 3536 ISOs: http://www.fsn.hu/?f=downloadcell.: +3630 306 6758 --- Cyrus Home Page: http://asg.web.cmu.edu/cyrus Cyrus Wiki/FAQ: http://cyruswiki.andrew.cmu.edu List Archives/Info: http://asg.web.cmu.edu/cyrus/mailing-list.html
Re: Paying for developers? (was: Re: Cyrus crashed on redundant platform - need better availability?)
On Tue, 14 Sep 2004, Attila Nagy wrote: So many users cried for this feature (to provide not just horizontal scalability with murder, but to have redundant backends which can hold each others replicas too) that I wonder: if it's so important to us, the cyrus users, why don't we collect some money and pass it to the developers? I wasn't following this entire thread, but if I'm not mistaken David Carter from the University of Cambridge already implemented what you're looking for: http://www-uxsup.csx.cam.ac.uk/~dpc22/cyrus/replication.html Maybe if you collect some money you could send it to him :) Thanks, Dave PGP/GPG Key: http://www.pitt.edu/~dgm/gpgkey.asc.txt --- Cyrus Home Page: http://asg.web.cmu.edu/cyrus Cyrus Wiki/FAQ: http://cyruswiki.andrew.cmu.edu List Archives/Info: http://asg.web.cmu.edu/cyrus/mailing-list.html
Re: Cyrus crashed on redundant platform - need better availability?
Paul Dekkers wrote: I'm not sure why the box crashed; there was nothing in the logs, there was nothing on the screen when we came there, it just booted up again. Of course I'm interested if anyone has any thoughts on this. Maybe it has nothing to do with your problem, but there is a timing issue with some intel xeon and p4 processors. Look at this HP advisory: http://tinyurl.com/63dxe even if it says that no field issues have been identified, I've experienced real random lock ups before updating the bios. Look if is there a bios update available from dell. Bye -- Luca Olivetti Wetron Automatización S.A. http://www.wetron.es/ Tel. +34 93 5883004 Fax +34 93 5883007 --- Cyrus Home Page: http://asg.web.cmu.edu/cyrus Cyrus Wiki/FAQ: http://cyruswiki.andrew.cmu.edu List Archives/Info: http://asg.web.cmu.edu/cyrus/mailing-list.html
Re: Cyrus crashed on redundant platform - need better availability?
Hi, Michael Loftis wrote: The theory only translates if you're using a JOURNALED file system. Linux ext3, reiserfs AIX JFS, Sun/others veritas are all examples of this. AFAIK FreeBSD hasn't any journalling file systems, Hmm, some say the use of softupdates preclude a journaling filesystem (see for instance http://kerneltrap.org/node.php?id=6). It's all a bit different with FreeBSD :-) That said, the machine shouldn't' have crashed in the first place, but you are running 5.x which is clearly labeled as *NOT* production (4.10 for that)... All of my produciton boxen are 4.x based (of the FreeBSD herd) You are right - 5.x is not stable yet, but 5.3 is very close to it. Since 5.3 is coming we thought it would be easier to install with 5.2.1 and upgrade to 5.3 rather then use 4.10 and upgrade that... And we might indeed face a 5.2.1-bug, that's why I mentioned the SMP, 3G and GEOM things, but might as well be something else with 5.x. Something else that was a vote for 5.x was the filesystem; 4.10 does not have UFS2. Apart from solving the issues we have with the machine I think we'd really look at the options for having redundany in application, as sketched in the High availability ... again subject :-) Maybe we'd install 5.3-BETA on the platform (I'll discuss it with another FreeBSD expert here :-)) Jure Pe?ar wrote: The only high availability i see here is the google way. Cyrus is offering you that with the 'murder' component. That's not really availability, but distributed risk. Exactly ... with murder taking care of keeping duplicated mailboxes in sync over a pool of backend machines (as i mentioned in the other mail), this would be perfect for all of us, i guess. Well, I don't know if it needs to be murder or some extension in the storage or something, but that's roughly my idea indead; synchronising two (or more? that's harder maybe) servers, just like doing an imapsync or rsync, but then... well, better! :-) (And without losing states and so forth.) The SPOF of the SCSI controller in the RAID box I'm willing to accept, but the filesystem is a bit harder. I'm curious what cyrus developers think of this, and I'm interested in what we can do to help. Paul --- Cyrus Home Page: http://asg.web.cmu.edu/cyrus Cyrus Wiki/FAQ: http://cyruswiki.andrew.cmu.edu List Archives/Info: http://asg.web.cmu.edu/cyrus/mailing-list.html
Re: Cyrus crashed on redundant platform - need better availability?
On Fri, 10 Sep 2004, Michael Loftis wrote: Date: Fri, 10 Sep 2004 13:15:05 -0600 From: Michael Loftis [EMAIL PROTECTED] To: Paul Dekkers [EMAIL PROTECTED], [EMAIL PROTECTED] Subject: Re: Cyrus crashed on redundant platform - need better availability? The theory only translates if you're using a JOURNALED file system. Linux ext3, reiserfs AIX JFS, Sun/others veritas are all examples of this. AFAIK FreeBSD hasn't any journalling file systems, i could be wrong though since I haven't really looked for one (my freebsd boxes just run...and run...and run...) That said, the machine shouldn't' have crashed in the first place, but you are running 5.x which is clearly labeled as *NOT* production (4.10 for that)... All of my produciton boxen are 4.x based (of the FreeBSD herd) However even a Journaled filesystem won't protect you completely from corruption. even the filesystems you list can loose data when there is a crash and if one system goes haywire and starts scribbling on the shared disk it will trash any filesystem. David Lang --On Friday, September 10, 2004 13:24 +0200 Paul Dekkers [EMAIL PROTECTED] wrote: Hi, We're implementing a new mailplatform running on two dell 2650-servers (2 xeon cpu's with each 3 Ghz, HTT and 3Gb of memory) and with a disk array of 4 Tb connected with a adaptec 39160 scsi controller for storage. We installed FreeBSD 5.2.1 on it, and - of course - cyrus 2.2.8 (from the ports) as IMAP server. Our MTA is postfix. There are two machines for redundancy. If one fails, the other one should take over: mount the disks from the array, and move on. Unfortunally, the primary server crashed twice already. The first time it did while synchronising two IMAP-spools from the old server to the new one. There was not much data on it back then. The second time was worse, around 10Gb of mail was stored on the disks. We discovered that the fsck took about 30 minutes, so although we have two machines for redundancy it takes still quite some time before the mail is available again. (And we still have about 90 Gb of mail to migrate, so when all users are migrated it takes much longer.) I mounted the filesystems synchronous now: although it slows down the system I hope it speeds up the fsck a bit when there is another crash. The second crash was while removing a lot of mailboxes (dm) while some of them where removed the same time using a webmail app (squirrelmail). I'm not sure why the box crashed; there was nothing in the logs, there was nothing on the screen when we came there, it just booted up again. Of course I'm interested if anyone has any thoughts on this. Although many on the list claim that this (having 2 boxes with 1 disk-array) is a nice way for redundancy I'm in doubt now if this is true. It still takes 30 mins before everything is back again! It seems to me that if there was a live version of cyrus available with a synchronised mail-spool, that there was no outage noticeable for users (except in losing a connection maybe). Am I right? Maybe it's time to continue on the High availability ... again-discussion we had a while ago. If the cyrus developers are able to implement this with some funding there are still some questions left for me: how much time would it take before a stable solution is ready? How many funding is expected? I still have to talk to management about this, but I would really support this development and I'm certainly willing to convince some managers. Regards, Paul --- Cyrus Home Page: http://asg.web.cmu.edu/cyrus Cyrus Wiki/FAQ: http://cyruswiki.andrew.cmu.edu List Archives/Info: http://asg.web.cmu.edu/cyrus/mailing-list.html -- Undocumented Features quote of the moment... It's not the one bullet with your name on it that you have to worry about; it's the twenty thousand-odd rounds labeled `occupant.' --Murphy's Laws of Combat --- Cyrus Home Page: http://asg.web.cmu.edu/cyrus Cyrus Wiki/FAQ: http://cyruswiki.andrew.cmu.edu List Archives/Info: http://asg.web.cmu.edu/cyrus/mailing-list.html -- There are two ways of constructing a software design. One way is to make it so simple that there are obviously no deficiencies. And the other way is to make it so complicated that there are no obvious deficiencies. -- C.A.R. Hoare --- Cyrus Home Page: http://asg.web.cmu.edu/cyrus Cyrus Wiki/FAQ: http://cyruswiki.andrew.cmu.edu List Archives/Info: http://asg.web.cmu.edu/cyrus/mailing-list.html
Re: Cyrus crashed on redundant platform - need better availability?
BTW -- if you want Stable (in case you didn't understand that from ym previous mail) go back to FreeBSD 4.x (say 4.10-STABLE or -SECURE) -- you've probably run into a platform bug, not a bug in Cyrus, since the whole machine went. --- Cyrus Home Page: http://asg.web.cmu.edu/cyrus Cyrus Wiki/FAQ: http://cyruswiki.andrew.cmu.edu List Archives/Info: http://asg.web.cmu.edu/cyrus/mailing-list.html
Re: Cyrus crashed on redundant platform - need better availability?
--On Friday, September 10, 2004 16:27 +0200 Paul Dekkers [EMAIL PROTECTED] wrote: What did the kernel improve? You are not using a clustered filesystem, right? RH kernels tend to coem up with bugs that noone else sees FYI (this is why my employer we're switching to Debian...) Well, it's UFS2 with softupdates, so yes. I'm afraid the journal was damaged in my case, there were serveral complaints while doing the fsck about softupdate inconsistencies. (The server crashed once more but since I mounted with -o sync now the fsck was much faster. I'll keep it that way for now untill we know what's really wrong - it was again with a large mail-folder synchronisation...) FWIW I can't call soft updates a journal. 9/10 times when i have had a crash, the soft updates journal either was corrupt, inconsistent, or made things worse. When running with soft updates many times I'd lose many days worth of mail on a restart. Hmm, I don't expect the problems to be SCSI-related. Maybe it has to do with GEOM and SMP in FreeBSD 5.2.1, but not the SCSI-bus itself. (There are two seperate controllers for both machines, they never see each other on the same SCSI bus...) Probably not, more likely something funkish in FBSD 5.2.1 I still think that it would be best to have two filesystems instead of one, so with mirroring on application level (cyrus)... :-) I tend to agree --- Cyrus Home Page: http://asg.web.cmu.edu/cyrus Cyrus Wiki/FAQ: http://cyruswiki.andrew.cmu.edu List Archives/Info: http://asg.web.cmu.edu/cyrus/mailing-list.html
Re: Cyrus crashed on redundant platform - need better availability?
On Fri, 10 Sep 2004 16:27:40 +0200 Paul Dekkers [EMAIL PROTECTED] wrote: Sebastian Hagedorn wrote: Right, works fine for us for the most part. Hasn't always been like that, but the most recent kernel updates by Red Hat have improved matters a lot. What did the kernel improve? You are not using a clustered filesystem, right? The kernel that shipped with RedHat AS 2.1 was useless for most of the tasks i tried it with. About three revisions later it became somewhat more usefull for non-oracle types of use, but i've rolled my own and am not following the state of it now. It's good but not perfect. We recently installed a huge SAN and are now in the process of moving over the mail data to reside there. Fibrechannel seems to be much more error tolerant than SCSI. I haven't had problems with the fiber itself, i've only had lots of fun with the firmware on the disks themselves and some with the qlogic drivers. I still think that it would be best to have two filesystems instead of one, so with mirroring on application level (cyrus)... :-) I'd rather see murder store a message on two sepparate machines ... Actually to have duplicated mailboxes in sync over a pool of backend machines, with murder taking care of backlogs when one of them would go down. -- Jure Pear --- Cyrus Home Page: http://asg.web.cmu.edu/cyrus Cyrus Wiki/FAQ: http://cyruswiki.andrew.cmu.edu List Archives/Info: http://asg.web.cmu.edu/cyrus/mailing-list.html
Re: Cyrus crashed on redundant platform - need better availability?
On Fri, 10 Sep 2004 16:32:33 +0200 Paul Dekkers [EMAIL PROTECTED] wrote: Hmm, then your fscks will run faster/with less problems, but there is still outage that you can prevent if there is failover in another way and availability/replication on the application level. If there are replicated spools it doesn't matter if the fsck takes long or not... although there will be a backlog of course. Yes, but right now there are no replicated spools on the app level so i'm doing the best i can as a sysadmin :) Is it possible to have an fsck running on one partition and have cyrus started already (so part of the mail-store, e.g. archives, is not available yet?) Not that i know ... i guess cyrus would be spewing lots of i/o errors back at you for the mailboxes that are on that fscking partition ;) The only high availability i see here is the google way. Cyrus is offering you that with the 'murder' component. That's not really availability, but distributed risk. Exactly ... with murder taking care of keeping duplicated mailboxes in sync over a pool of backend machines (as i mentioned in the other mail), this would be perfect for all of us, i guess. BTW, you're mentioning FreeBSD ... doesn't it have some sort of background fsck while the filesystem is moutned rw? It can, but I'm not sure if that's what I prefer. I'm not sure how mature it is with FreeBSD, and I prefer to have mail-integrety over a quick restore. I can't speak about maturity of a certain FreeBSD component as i'm a linux guy, but what i hear it should just work. -- Jure Pear --- Cyrus Home Page: http://asg.web.cmu.edu/cyrus Cyrus Wiki/FAQ: http://cyruswiki.andrew.cmu.edu List Archives/Info: http://asg.web.cmu.edu/cyrus/mailing-list.html
Re: Cyrus crashed on redundant platform - need better availability?
The theory only translates if you're using a JOURNALED file system. Linux ext3, reiserfs AIX JFS, Sun/others veritas are all examples of this. AFAIK FreeBSD hasn't any journalling file systems, i could be wrong though since I haven't really looked for one (my freebsd boxes just run...and run...and run...) That said, the machine shouldn't' have crashed in the first place, but you are running 5.x which is clearly labeled as *NOT* production (4.10 for that)... All of my produciton boxen are 4.x based (of the FreeBSD herd) --On Friday, September 10, 2004 13:24 +0200 Paul Dekkers [EMAIL PROTECTED] wrote: Hi, We're implementing a new mailplatform running on two dell 2650-servers (2 xeon cpu's with each 3 Ghz, HTT and 3Gb of memory) and with a disk array of 4 Tb connected with a adaptec 39160 scsi controller for storage. We installed FreeBSD 5.2.1 on it, and - of course - cyrus 2.2.8 (from the ports) as IMAP server. Our MTA is postfix. There are two machines for redundancy. If one fails, the other one should take over: mount the disks from the array, and move on. Unfortunally, the primary server crashed twice already. The first time it did while synchronising two IMAP-spools from the old server to the new one. There was not much data on it back then. The second time was worse, around 10Gb of mail was stored on the disks. We discovered that the fsck took about 30 minutes, so although we have two machines for redundancy it takes still quite some time before the mail is available again. (And we still have about 90 Gb of mail to migrate, so when all users are migrated it takes much longer.) I mounted the filesystems synchronous now: although it slows down the system I hope it speeds up the fsck a bit when there is another crash. The second crash was while removing a lot of mailboxes (dm) while some of them where removed the same time using a webmail app (squirrelmail). I'm not sure why the box crashed; there was nothing in the logs, there was nothing on the screen when we came there, it just booted up again. Of course I'm interested if anyone has any thoughts on this. Although many on the list claim that this (having 2 boxes with 1 disk-array) is a nice way for redundancy I'm in doubt now if this is true. It still takes 30 mins before everything is back again! It seems to me that if there was a live version of cyrus available with a synchronised mail-spool, that there was no outage noticeable for users (except in losing a connection maybe). Am I right? Maybe it's time to continue on the High availability ... again-discussion we had a while ago. If the cyrus developers are able to implement this with some funding there are still some questions left for me: how much time would it take before a stable solution is ready? How many funding is expected? I still have to talk to management about this, but I would really support this development and I'm certainly willing to convince some managers. Regards, Paul --- Cyrus Home Page: http://asg.web.cmu.edu/cyrus Cyrus Wiki/FAQ: http://cyruswiki.andrew.cmu.edu List Archives/Info: http://asg.web.cmu.edu/cyrus/mailing-list.html -- Undocumented Features quote of the moment... It's not the one bullet with your name on it that you have to worry about; it's the twenty thousand-odd rounds labeled `occupant.' --Murphy's Laws of Combat --- Cyrus Home Page: http://asg.web.cmu.edu/cyrus Cyrus Wiki/FAQ: http://cyruswiki.andrew.cmu.edu List Archives/Info: http://asg.web.cmu.edu/cyrus/mailing-list.html
Re: Cyrus crashed on redundant platform - need better availability?
On Fri, 10 Sep 2004 13:24:42 +0200 Paul Dekkers [EMAIL PROTECTED] wrote: Although many on the list claim that this (having 2 boxes with 1 disk-array) is a nice way for redundancy I'm in doubt now if this is true. It still takes 30 mins before everything is back again! It seems to me that if there was a live version of cyrus available with a synchronised mail-spool, that there was no outage noticeable for users (except in losing a connection maybe). Am I right? Having 2 boxes with one disk array leaves you wit a single point of failure that you wouldn't think of immediately: filesystem. I learned that the hard way. I'm planning to 'redesign' our storage: instead of one big volume that fscks for hours, i'm going to split in in many mirrors and use them as cyrus partitions. This way they could all fsck in parrallel. I'm going to lose the 'single instance store' capability, but thats a tradeoff that i'm willing to take. It happened to me at least once that the machine that crashed corrupted the filesystem in a way that the machine that took over also crashed within hours... Maybe it's time to continue on the High availability ... again-discussion we had a while ago. If the cyrus developers are able to implement this with some funding there are still some questions left for me: how much time would it take before a stable solution is ready? How many funding is expected? I still have to talk to management about this, but I would really support this development and I'm certainly willing to convince some managers. The only high availability i see here is the google way. Cyrus is offering you that with the 'murder' component. BTW, you're mentioning FreeBSD ... doesn't it have some sort of background fsck while the filesystem is moutned rw? -- Jure Pear --- Cyrus Home Page: http://asg.web.cmu.edu/cyrus Cyrus Wiki/FAQ: http://cyruswiki.andrew.cmu.edu List Archives/Info: http://asg.web.cmu.edu/cyrus/mailing-list.html
Re: Cyrus crashed on redundant platform - need better availability?
Hi, --On Freitag, 10. September 2004 13:24 Uhr +0200 Paul Dekkers [EMAIL PROTECTED] wrote: We're implementing a new mailplatform running on two dell 2650-servers (2 xeon cpu's with each 3 Ghz, HTT and 3Gb of memory) and with a disk array of 4 Tb connected with a adaptec 39160 scsi controller for storage. We installed FreeBSD 5.2.1 on it, and - of course - cyrus 2.2.8 (from the ports) as IMAP server. Our MTA is postfix. that's similar to our setup, be we are currently running Red Hat Advanced Server 2.1, Cyrus 2.1.16 and sendmail. There are two machines for redundancy. If one fails, the other one should take over: mount the disks from the array, and move on. Right, works fine for us for the most part. Hasn't always been like that, but the most recent kernel updates by Red Hat have improved matters a lot. Unfortunally, the primary server crashed twice already. The first time it did while synchronising two IMAP-spools from the old server to the new one. There was not much data on it back then. The second time was worse, around 10Gb of mail was stored on the disks. We discovered that the fsck took about 30 minutes, Isn't your filesystem journaled? We use ext3 for ours. There *have* been a few occasions where the journal had been damaged as well (forcing us to run fsck), but those have been few and far between. In all other instances the failover is nearly instantaneous. Although many on the list claim that this (having 2 boxes with 1 disk-array) is a nice way for redundancy I'm in doubt now if this is true. It's good but not perfect. We recently installed a huge SAN and are now in the process of moving over the mail data to reside there. Fibrechannel seems to be much more error tolerant than SCSI. Cheers, Sebastian Hagedorn -- Sebastian Hagedorn M.A. - RZKR-R1 (Gebäude 52), Zimmer 18 Zentrum für angewandte Informatik - Universitätsweiter Service RRZK Universität zu Köln / Cologne University - Tel. +49-221-478-5587 pgprNRdYpcyzG.pgp Description: PGP signature
Re: Cyrus crashed on redundant platform - need better availability?
Hi, Sebastian Hagedorn wrote: There are two machines for redundancy. If one fails, the other one should take over: mount the disks from the array, and move on. Right, works fine for us for the most part. Hasn't always been like that, but the most recent kernel updates by Red Hat have improved matters a lot. What did the kernel improve? You are not using a clustered filesystem, right? Unfortunally, the primary server crashed twice already. The first time it did while synchronising two IMAP-spools from the old server to the new one. There was not much data on it back then. The second time was worse, around 10Gb of mail was stored on the disks. We discovered that the fsck took about 30 minutes, Isn't your filesystem journaled? We use ext3 for ours. There *have* been a few occasions where the journal had been damaged as well (forcing us to run fsck), but those have been few and far between. In all other instances the failover is nearly instantaneous. Well, it's UFS2 with softupdates, so yes. I'm afraid the journal was damaged in my case, there were serveral complaints while doing the fsck about softupdate inconsistencies. (The server crashed once more but since I mounted with -o sync now the fsck was much faster. I'll keep it that way for now untill we know what's really wrong - it was again with a large mail-folder synchronisation...) Although many on the list claim that this (having 2 boxes with 1 disk-array) is a nice way for redundancy I'm in doubt now if this is true. It's good but not perfect. We recently installed a huge SAN and are now in the process of moving over the mail data to reside there. Fibrechannel seems to be much more error tolerant than SCSI. Hmm, I don't expect the problems to be SCSI-related. Maybe it has to do with GEOM and SMP in FreeBSD 5.2.1, but not the SCSI-bus itself. (There are two seperate controllers for both machines, they never see each other on the same SCSI bus...) I still think that it would be best to have two filesystems instead of one, so with mirroring on application level (cyrus)... :-) Paul --- Cyrus Home Page: http://asg.web.cmu.edu/cyrus Cyrus Wiki/FAQ: http://cyruswiki.andrew.cmu.edu List Archives/Info: http://asg.web.cmu.edu/cyrus/mailing-list.html
Re: Cyrus crashed on redundant platform - need better availability?
Jure Pear wrote: Although many on the list claim that this (having 2 boxes with 1 disk-array) is a nice way for redundancy I'm in doubt now if this is true. It still takes 30 mins before everything is back again! It seems to me that if there was a live version of cyrus available with a synchronised mail-spool, that there was no outage noticeable for users (except in losing a connection maybe). Am I right? Having 2 boxes with one disk array leaves you wit a single point of failure that you wouldn't think of immediately: filesystem. I learned that the hard way. Yes, I agree. I'm planning to 'redesign' our storage: instead of one big volume that fscks for hours, i'm going to split in in many mirrors and use them as cyrus partitions. This way they could all fsck in parrallel. I'm going to lose the 'single instance store' capability, but thats a tradeoff that i'm willing to take. Hmm, then your fscks will run faster/with less problems, but there is still outage that you can prevent if there is failover in another way and availability/replication on the application level. If there are replicated spools it doesn't matter if the fsck takes long or not... although there will be a backlog of course. Is it possible to have an fsck running on one partition and have cyrus started already (so part of the mail-store, e.g. archives, is not available yet?) It happened to me at least once that the machine that crashed corrupted the filesystem in a way that the machine that took over also crashed within hours... Maybe it's time to continue on the High availability ... again-discussion we had a while ago. If the cyrus developers are able to implement this with some funding there are still some questions left for me: how much time would it take before a stable solution is ready? How many funding is expected? I still have to talk to management about this, but I would really support this development and I'm certainly willing to convince some managers. The only high availability i see here is the google way. Cyrus is offering you that with the 'murder' component. That's not really availability, but distributed risk. BTW, you're mentioning FreeBSD ... doesn't it have some sort of background fsck while the filesystem is moutned rw? It can, but I'm not sure if that's what I prefer. I'm not sure how mature it is with FreeBSD, and I prefer to have mail-integrety over a quick restore. Paul --- Cyrus Home Page: http://asg.web.cmu.edu/cyrus Cyrus Wiki/FAQ: http://cyruswiki.andrew.cmu.edu List Archives/Info: http://asg.web.cmu.edu/cyrus/mailing-list.html