Yes simple is a good thing, but you have to decide for yourself what risk and downtime is acceptable, how long to recover and how many writes is acceptable to lose.
E.g. Is a one daily backup acceptable for recovery (and you lose changes since then), or do you need every single write to the hour/minute ?
The more continously/frequently you write to both nodes, the more likely are sync errors, and also less recoverability of your bdb logs in the event of catastrophic power failure or inproper shutdown.
I dont think anyone can decide this for you.. You need to evaluate your level of risk and what loss / downtime is acceptable to you / your employer / your customers etc.,
Cheers Brett
-----Original Message----- From: Hideo NAKAMITSU nomo@bluecoara.net Sent: Monday, 29 June 2009 11:51 AM To: openldap-technical@openldap.org Subject: Re: MirrorMode VS load balancing
and how about 1write/sec, 10writes/sec, 50writes/sec, 100writes/sec... I want to know this threshold like,
- if writes happen 0-5/sec, it's 99% safe (at the worst unlucky case, only 2writes at the same time may break something) - 6-50/sec, may safe or not. - more than 50/sec, danger.
Why not use a Multi-Master replication for this situation ?
Regards, Jakjr
On Tue, Jun 30, 2009 at 6:58 AM, Brett Maxfield brett.maxfield@gmail.comwrote:
Yes simple is a good thing, but you have to decide for yourself what risk and downtime is acceptable, how long to recover and how many writes is acceptable to lose.
E.g. Is a one daily backup acceptable for recovery (and you lose changes since then), or do you need every single write to the hour/minute ?
The more continously/frequently you write to both nodes, the more likely are sync errors, and also less recoverability of your bdb logs in the event of catastrophic power failure or inproper shutdown.
I dont think anyone can decide this for you.. You need to evaluate your level of risk and what loss / downtime is acceptable to you / your employer / your customers etc.,
Cheers Brett
-----Original Message----- From: Hideo NAKAMITSU nomo@bluecoara.net Sent: Monday, 29 June 2009 11:51 AM To: openldap-technical@openldap.org Subject: Re: MirrorMode VS load balancing
and how about 1write/sec, 10writes/sec, 50writes/sec, 100writes/sec... I want to know this threshold like,
- if writes happen 0-5/sec, it's 99% safe
(at the worst unlucky case, only 2writes at the same time may break something)
- 6-50/sec, may safe or not.
- more than 50/sec, danger.
On Tue, Jun 30, 2009 at 10:19 PM, jakjr joao.alfredo@gmail.com wrote:
Why not use a Multi-Master replication for this situation ?
I don't propose one or the other. But here are some thoughts against :
http://www.watersprings.org/pub/id/draft-zeilenga-ldup-harmful-02.txt
The core issue for me seems to be conflicts with determining who has the "newest" data, and efficiency of transfer. If replication is predominantly in one direction, the replication works better with less latency and there is less opportunity for conflicts.
Mirrormode and multimaster are configured the same in openldap, as i understand it, except it is called mirrormode when you primarily write through one node most of the time.
In mirrormode both nodes keep a full copy of the DIT so it has the benefits of full redundancy, but without the possibility where both servers get an update at the same time for the same data, and cannot resolve the conflict themselves (in which case one server will be wrong).
There is also the side benefit / last resort of having two seperate sets of BDB transaction logs, to help recover consistancy in the event of one serious server failure or bdb corruption due to improper power / shutdowns etc.,
Cheers Brett
Brett @Google wrote:
On Tue, Jun 30, 2009 at 10:19 PM, jakjr <joao.alfredo@gmail.com mailto:joao.alfredo@gmail.com> wrote:
Why not use a Multi-Master replication for this situation ?I don't propose one or the other. But here are some thoughts against :
http://www.watersprings.org/pub/id/draft-zeilenga-ldup-harmful-02.txt
The core issue for me seems to be conflicts with determining who has the "newest" data, and efficiency of transfer. If replication is predominantly in one direction, the replication works better with less latency and there is less opportunity for conflicts.
Well, even with simple master-slave replication there can be subtle problems if an application writes to the master (even through slapo-chain) and reads immediately from the slave.
One has to carefully examine the behaviour of the LDAP client applications before making decisions on replication and fail-over scenarios.
Ciao, Michael.
Hi, Brett,
On Tue, 30 Jun 2009 19:58:07 +1000 Brett Maxfield brett.maxfield@gmail.com wrote:
Yes simple is a good thing, but you have to decide for yourself what risk and downtime is acceptable, how long to recover and how many writes is acceptable to lose.
E.g. Is a one daily backup acceptable for recovery (and you lose changes since then), or do you need every single write to the hour/minute ?
The more continously/frequently you write to both nodes, the more likely are sync errors, and also less recoverability of your bdb logs in the event of catastrophic power failure or inproper shutdown.
I dont think anyone can decide this for you.. You need to evaluate your level of risk and what loss / downtime is acceptable to you / your employer / your customers etc.,
Thanks for your impression.
actually, I was thinking it's very difficult to answer YES or NO. even if there are few writes, it may cause database corruption at the worst case.
it's not 100% safe, but in most cases, must be 99.99% safe. but in case of failure, it's important to monitor every nodes, and if the difference is found, administrator must fix it by hands from backup or such.
anyway, it was good opportunity to know not only me but also your thoughts.
Regards,
openldap-technical@openldap.org
-
Brett ＠Google -
Brett Maxfield
-
Hideo NAKAMITSU -
jakjr -
Michael Ströder