Re: [SunRay-Users] Failover group vs. simple DNS registers
On 2014-06-19 22:37, Meik Hellmund wrote: Of course all the servers in a failover group have to provide the same home directories (via NFS, usually). This is standard unix practice and has nothing to do with the SRS software. So, the home is not replicated. It is simply always the same home. +1 The session in this context reflects more the active resources (RAM consumed by running processes) related to a particular user. When the terminal disconnects, the user's desktop, browser and other programs remain running. Thanks to a FOG, if a user later reconnects to any server initially (due to round-robin) and "his" session is found on any server in the FOG, the terminal will be redirected to be served by the same server and session as before. Note that a similar effect can also be scripted with "utswitch" (IIRC) for example to dedicate some uncrowded servers to some particular users (tokens) for responsiveness and similar reasons. HTH, //Jim ___ SunRay-Users mailing list SunRay-Users@filibeto.org http://www.filibeto.org/mailman/listinfo/sunray-users
Re: [SunRay-Users] Failover group vs. simple DNS registers
On Thu, 19 Jun 2014 21:03:41 +0100 Edwin Marqe wrote: > Thanks very much for your response. I will definitely follow your advices, > they even are requirements for our organization so I have no choice anyway > :-) > > My original mail was motivated more like due to the concept of "session" > here - In the SRS administration guide the session is defined as a set of > assigned resources to an user. It's also defined (from the user's point of > view) as a desktop session. As Bjoern said, once a server loses its > connectivity, those sessions are lost. > > However, I've seen in a bunch of places that stated that even these > sessions are lost, once the users reconnect, they'll recover the state of > their desktop session (meaning the same desktop environment, the same > windows, applications, etc.). So that's what still doesn't fit to me - Most > of this data is stored in the user's home directory (under .gconf, > .local... directories). So if by default homes are not replicated, how can > the SRSS failover group know each user's desktop state? Of course all the servers in a failover group have to provide the same home directories (via NFS, usually). This is standard unix practice and has nothing to do with the SRS software. So, the home is not replicated. It is simply always the same home. regards, Meik -- Meik Hellmund Mathematisches Institut, Uni Leipzig e-mail: meik.hellm...@math.uni-leipzig.de ___ SunRay-Users mailing list SunRay-Users@filibeto.org http://www.filibeto.org/mailman/listinfo/sunray-users
Re: [SunRay-Users] Failover group vs. simple DNS registers
Thanks very much for your response. I will definitely follow your advices, they even are requirements for our organization so I have no choice anyway :-) My original mail was motivated more like due to the concept of "session" here - In the SRS administration guide the session is defined as a set of assigned resources to an user. It's also defined (from the user's point of view) as a desktop session. As Bjoern said, once a server loses its connectivity, those sessions are lost. However, I've seen in a bunch of places that stated that even these sessions are lost, once the users reconnect, they'll recover the state of their desktop session (meaning the same desktop environment, the same windows, applications, etc.). So that's what still doesn't fit to me - Most of this data is stored in the user's home directory (under .gconf, .local... directories). So if by default homes are not replicated, how can the SRSS failover group know each user's desktop state? Well, all that assuming they actually recover that desktop session and the places where stated the desktop environment will be recovered are right. If not, I still don't see a big advantage between using a failover group and simply reconnecting to a new server chosen by DNS beyond the balancing algorithm. 2014-06-16 21:31 GMT+01:00 Jim Klimov : > On 2014-06-15 19:48, Edwin Marqe wrote: > >> Thanks Bjoern. Does your last point mean that after the client >> reconnects it would recover the last known state it had on the failed >> server, meaning i.e. the same windows opened, same desktop configuration >> etc? In this case, would that mean that there's also a 'home' >> sincronization between the failover group servers? >> > > Agreeing with Bjoern, I want to address the last sentence: > do you mean a home directory here, and their synchronization > among Sun Ray servers? > > With any sufficiently large installation (and having several > terminal servers alone qualifies for that, IMHO) you should > really look towards a centralized networking infrastructure, > which includes home directories residing on an NFS server and > user accounts (especially non-kiosk ones) defined in the LDAP > catalogs (Sun DSEE, Oracle DSEE, ForgeRock OpenDJ, maybe even > OpenLDAP or MS Active Directory with Unix extensions). For > more resilience, you might replicate the LDAP catalogs, maybe > keeping a copy on each server and preferring it to the other > copies (search order, which can be defined in LDAP client > profiles stored in the catalog, and you assign each server > to its own profile). > > While it is discouraged by documentation, there are tricks to > allow a server hosting the LDAP service to be its own client. > > As one of the attributes in LDAP accounts you can specify the > home directories (though it is usually /home/username for all) > and also you can define automounter maps for autofs (so that > if you have several NFS servers hosting different sets of > users - these relations can be defined here). > > For failover of the homedir storage you use any number of > solutions, but should really prefer those based on OpenZFS > with illumos or BSD kernels (commercial or free, though the > reliable/HA ones tend to be more expensive in gear and paying > a price to one of the many developer companies for software > license and support is reasonable to keep the setup running). > In particular, for NFS and ZFS you should look towards systems > with dedicated ZIL (write-logging) devices preferably with DDR > RAM for active storage rather than SSDs or HDDs, since all NFS > write I/O is synchronous by design and lags greatly on ordinary > HDDs (must be really committed to storage before ACKing, and > servers that don't do this and seem fast - they lie and are an > accident waiting to happen in an emergency poweroff, kernel > crash, etc. which would not let them flush the caches for the > delayed writes). > > HTH, > Jim Klimov > > > ___ > SunRay-Users mailing list > SunRay-Users@filibeto.org > http://www.filibeto.org/mailman/listinfo/sunray-users > ___ SunRay-Users mailing list SunRay-Users@filibeto.org http://www.filibeto.org/mailman/listinfo/sunray-users
Re: [SunRay-Users] Failover group vs. simple DNS registers
On 2014-06-15 19:48, Edwin Marqe wrote: Thanks Bjoern. Does your last point mean that after the client reconnects it would recover the last known state it had on the failed server, meaning i.e. the same windows opened, same desktop configuration etc? In this case, would that mean that there's also a 'home' sincronization between the failover group servers? Agreeing with Bjoern, I want to address the last sentence: do you mean a home directory here, and their synchronization among Sun Ray servers? With any sufficiently large installation (and having several terminal servers alone qualifies for that, IMHO) you should really look towards a centralized networking infrastructure, which includes home directories residing on an NFS server and user accounts (especially non-kiosk ones) defined in the LDAP catalogs (Sun DSEE, Oracle DSEE, ForgeRock OpenDJ, maybe even OpenLDAP or MS Active Directory with Unix extensions). For more resilience, you might replicate the LDAP catalogs, maybe keeping a copy on each server and preferring it to the other copies (search order, which can be defined in LDAP client profiles stored in the catalog, and you assign each server to its own profile). While it is discouraged by documentation, there are tricks to allow a server hosting the LDAP service to be its own client. As one of the attributes in LDAP accounts you can specify the home directories (though it is usually /home/username for all) and also you can define automounter maps for autofs (so that if you have several NFS servers hosting different sets of users - these relations can be defined here). For failover of the homedir storage you use any number of solutions, but should really prefer those based on OpenZFS with illumos or BSD kernels (commercial or free, though the reliable/HA ones tend to be more expensive in gear and paying a price to one of the many developer companies for software license and support is reasonable to keep the setup running). In particular, for NFS and ZFS you should look towards systems with dedicated ZIL (write-logging) devices preferably with DDR RAM for active storage rather than SSDs or HDDs, since all NFS write I/O is synchronous by design and lags greatly on ordinary HDDs (must be really committed to storage before ACKing, and servers that don't do this and seem fast - they lie and are an accident waiting to happen in an emergency poweroff, kernel crash, etc. which would not let them flush the caches for the delayed writes). HTH, Jim Klimov ___ SunRay-Users mailing list SunRay-Users@filibeto.org http://www.filibeto.org/mailman/listinfo/sunray-users
Re: [SunRay-Users] Failover group vs. simple DNS registers
> Thanks Bjoern. Does your last point mean that after the client reconnects it > would recover the last known state it had on the failed server, meaning i.e. > the same windows opened, same desktop configuration etc? No, if a server fails, all session state is lost. But If your client reconnects (think powercycle) then with proper failover group setup this client would get redirected to the original server and could continue the session there. cheers bjoern ___ SunRay-Users mailing list SunRay-Users@filibeto.org http://www.filibeto.org/mailman/listinfo/sunray-users
Re: [SunRay-Users] Failover group vs. simple DNS registers
Thanks Bjoern. Does your last point mean that after the client reconnects it would recover the last known state it had on the failed server, meaning i.e. the same windows opened, same desktop configuration etc? In this case, would that mean that there's also a 'home' sincronization between the failover group servers? Thanks so much 2014-06-15 18:31 GMT+01:00 Bjoern Rost : > > So here goes the question: Then what's the benefit of configuring a > failover group? I mean, it's ok that the servers share session information > for the users whose server has fell off, but what's the advantage of this > approach if the session for the user is interrupted and they have to > reconnect? > > - you can set a server to not accept new sessions and wait for existing > sessions to drain out before performing maintenance > - failover groups also do load-balancing for new connections > - when a client reconnects and has an existing session, it will get > redirected there. unlike “just” DNS RR where it could happen that one > client has multiple sessions on multiple servers > > cheers > Bjoern > > ___ > SunRay-Users mailing list > SunRay-Users@filibeto.org > http://www.filibeto.org/mailman/listinfo/sunray-users > ___ SunRay-Users mailing list SunRay-Users@filibeto.org http://www.filibeto.org/mailman/listinfo/sunray-users
Re: [SunRay-Users] Failover group vs. simple DNS registers
> So here goes the question: Then what's the benefit of configuring a failover > group? I mean, it's ok that the servers share session information for the > users whose server has fell off, but what's the advantage of this approach if > the session for the user is interrupted and they have to reconnect? - you can set a server to not accept new sessions and wait for existing sessions to drain out before performing maintenance - failover groups also do load-balancing for new connections - when a client reconnects and has an existing session, it will get redirected there. unlike “just” DNS RR where it could happen that one client has multiple sessions on multiple servers cheers Bjoern ___ SunRay-Users mailing list SunRay-Users@filibeto.org http://www.filibeto.org/mailman/listinfo/sunray-users
[SunRay-Users] Failover group vs. simple DNS registers
Hi list, We've been reading documentation for SRS 5.4 relative to Failover group configuration. As per the documentation, when a client tries to discover possible servers to connect, DNS record for 'sunray-servers' is looked up and picked one randomly. If I understood it correctly, on the failover group part, seems that when one of the servers of the group fails, the sessions are distributed amongst the remaining servers, but that doesn't mean that the session is redirected. Instead, the client will have to reconnect and therefore will fall in any of the remaining servers, as the DNS record will be looked up again. So here goes the question: Then what's the benefit of configuring a failover group? I mean, it's ok that the servers share session information for the users whose server has fell off, but what's the advantage of this approach if the session for the user is interrupted and they have to reconnect? Any hints will be very helpful! Thanks! Regards, Edwin ___ SunRay-Users mailing list SunRay-Users@filibeto.org http://www.filibeto.org/mailman/listinfo/sunray-users
