Full_Name: Ryan Steele Version: 2.4.18 OS: Ubuntu Server URL: ftp://ftp.openldap.org/incoming/ Submission from: (NULL) (207.106.239.81) Per conversation on the #openldap IRC channel on Freenode with Howard Chu, it has been deemed appropriate to modify the behavior of updateref to answer referrals on the backend. This would allow one to use updateref to automatically chase referrals for individual backends, instead of it being all or nothing. Here is the tail end of the channel conversation: <hyc> when you configure an updateref on a backend, that referral is only generated in the frontend so putting the overlay on the database, misses it... <rgsteele> Well, I am intending to set an updateref on that backend <hyc> then you have a problem <rgsteele> Hm, so you can't do automatic referrals on individual backends then? <rgsteele> That was my original question, or at least the intent of it. <hyc> not using updateref, no <rgsteele> Does your response imply there's another way? <hyc> probably we should fix updateref to be generated at the backend level <rgsteele> That would be good - I'd be happy to write tests or something to help with that. <hyc> the "other way" is the normal referral mechanism - using referral entries inside a database <hyc> but like I said, updateref is a relic, we've carried it forward without really adapting it <hyc> probably should file an ITS about this <rgsteele> Interesting, I wasn't aware of that method. I'll have to do some research on that - thanks! Also, is there anything I can do to help get the process of fixing updateref to generate referrals on the backend? <rgsteele> I can file the ITS if you like? <hyc> go ahead
> Full_Name: Ryan Steele > Version: 2.4.18 > OS: Ubuntu Server > URL: ftp://ftp.openldap.org/incoming/ > Submission from: (NULL) (207.106.239.81) > > > Per conversation on the #openldap IRC channel on Freenode with Howard Chu, > it > has been deemed appropriate to modify the behavior of updateref to answer > referrals on the backend. This would allow one to use updateref to > automatically chase referrals for individual backends, instead of it being > all > or nothing. Here is the tail end of the channel conversation: > > > <hyc> when you configure an updateref on a backend, that referral is only > generated in the frontend so putting the overlay on the database, misses > it... > > <rgsteele> Well, I am intending to set an updateref on that backend > > <hyc> then you have a problem > > <rgsteele> Hm, so you can't do automatic referrals on individual backends > then? > > <rgsteele> That was my original question, or at least the intent of it. > > <hyc> not using updateref, no > > <rgsteele> Does your response imply there's another way? > > <hyc> probably we should fix updateref to be generated at the backend > level > > <rgsteele> That would be good - I'd be happy to write tests or something > to help > with that. > > <hyc> the "other way" is the normal referral mechanism - using referral > entries > inside a database > > <hyc> but like I said, updateref is a relic, we've carried it forward > without > really adapting it > > <hyc> probably should file an ITS about this > > <rgsteele> Interesting, I wasn't aware of that method. I'll have to do > some > research on that - thanks! Also, is there anything I can do to help get > the > process of fixing updateref to generate referrals on the backend? > > <rgsteele> I can file the ITS if you like? > > <hyc> go ahead It is correct that referrals are generated by the frontend, but the frontend uses information contained in the updateref of each database. The slapo-chain needs to be global, in order to intercept those referrals. Although redesigning referral generation in slapd could streamline things, I don't quite see what issues can't be addressed using a global instance of slapo-chain. You can configure referral chasing differently for different domains using the chain-uri directive. I favor fixing only what's actually broken. p.
masarati@aero.polimi.it wrote: >> Full_Name: Ryan Steele >> Version: 2.4.18 >> OS: Ubuntu Server >> URL: ftp://ftp.openldap.org/incoming/ >> Submission from: (NULL) (207.106.239.81) >> >> >> Per conversation on the #openldap IRC channel on Freenode with Howard Chu, >> it >> has been deemed appropriate to modify the behavior of updateref to answer >> referrals on the backend. This would allow one to use updateref to >> automatically chase referrals for individual backends, instead of it being >> all >> or nothing. Here is the tail end of the channel conversation: >> >> >> <hyc> when you configure an updateref on a backend, that referral is only >> generated in the frontend so putting the overlay on the database, misses >> it... >> >> <rgsteele> Well, I am intending to set an updateref on that backend >> >> <hyc> then you have a problem >> >> <rgsteele> Hm, so you can't do automatic referrals on individual backends >> then? >> >> <rgsteele> That was my original question, or at least the intent of it. >> >> <hyc> not using updateref, no >> >> <rgsteele> Does your response imply there's another way? >> >> <hyc> probably we should fix updateref to be generated at the backend >> level >> >> <rgsteele> That would be good - I'd be happy to write tests or something >> to help >> with that. >> >> <hyc> the "other way" is the normal referral mechanism - using referral >> entries >> inside a database >> >> <hyc> but like I said, updateref is a relic, we've carried it forward >> without >> really adapting it >> >> <hyc> probably should file an ITS about this >> >> <rgsteele> Interesting, I wasn't aware of that method. I'll have to do >> some >> research on that - thanks! Also, is there anything I can do to help get >> the >> process of fixing updateref to generate referrals on the backend? >> >> <rgsteele> I can file the ITS if you like? >> >> <hyc> go ahead > > It is correct that referrals are generated by the frontend, but the > frontend uses information contained in the updateref of each database. > The slapo-chain needs to be global, in order to intercept those referrals. > > Although redesigning referral generation in slapd could streamline things, > I don't quite see what issues can't be addressed using a global instance > of slapo-chain. You can configure referral chasing differently for > different domains using the chain-uri directive. I favor fixing only > what's actually broken. The request revolves around having different behavior for different databases. With a global slapo-chain you cannot turn off referral chasing; URIs that are not explicitly configured will still be chased. Also in terms of logical abstraction the current behavior is wrong; the frontend should of course generate the default_referral responses but updateref is a per-backend item and should be generated at the backend layer. Ideally this would all have been fixed to behave correctly by moving all replication support into an overlay, where it properly belongs. As an interim step it would be trivial to write an updateref overlay that supersedes the current updateref behavior. -- -- Howard Chu CTO, Symas Corp. http://www.symas.com Director, Highland Sun http://highlandsun.com/hyc/ Chief Architect, OpenLDAP http://www.openldap.org/project/
[snip] >> Although redesigning referral generation in slapd could streamline >> things, >> I don't quite see what issues can't be addressed using a global instance >> of slapo-chain. You can configure referral chasing differently for >> different domains using the chain-uri directive. I favor fixing only >> what's actually broken. > > The request revolves around having different behavior for different > databases. > With a global slapo-chain you cannot turn off referral chasing; URIs that > are > not explicitly configured will still be chased. > > Also in terms of logical abstraction the current behavior is wrong; the > frontend should of course generate the default_referral responses but > updateref is a per-backend item and should be generated at the backend > layer. > Ideally this would all have been fixed to behave correctly by moving all > replication support into an overlay, where it properly belongs. As an > interim > step it would be trivial to write an updateref overlay that supersedes the > current updateref behavior. As I said, I recognize the usefulness of such a change: I first moved replication into an overlay (ITS#4261). I'm questioning its need for the case at hand. Perhaps allowing slapo-chain to ignore some URIs would be easier to implement, and generally useful in other applications. In any case, I might spare some time to revitalize (ITS#4261), now that slurpd is no longer around and thus things simplify significantly. p.
On 04/25/2010 12:49 AM, hyc@symas.com wrote: > Also in terms of logical abstraction the current behavior is wrong; the > frontend should of course generate the default_referral responses but > updateref is a per-backend item and should be generated at the backend layer. > Ideally this would all have been fixed to behave correctly by moving all > replication support into an overlay, where it properly belongs. As an interim > step it would be trivial to write an updateref overlay that supersedes the > current updateref behavior. Updateref processing should imo be kept out of any replication modules, it is needed in databases where syncrepl isn't configured too. I'm using it on my subordinate databases (with an artificial updatedn so that slapd accepts it..) when syncrepl is configured on the superior glue database without managing all of the subordinate databases. I.e, the infamous test058... I was having the same "chain only on some of the databases" requirement and is using a patch that allows it. I consider this patch more of a hack than a correct solution, which is why I haven't contributed it. A separate updateref overlay apear to me as the best solution. But for what it's worth, my patch is now in: ftp://ftp.openldap.org/incoming/rein-ITS6531.patch An advantage with this patch over a new overlay is that it should be usable without any configuration changes (provided the chain overlay have already been configured in the databases and found to not work as expected there that is ... ;-). Rein
Rein Tollevik wrote: > On 04/25/2010 12:49 AM, hyc@symas.com wrote: > >> Also in terms of logical abstraction the current behavior is wrong; the >> frontend should of course generate the default_referral responses but >> updateref is a per-backend item and should be generated at the backend layer. >> Ideally this would all have been fixed to behave correctly by moving all >> replication support into an overlay, where it properly belongs. As an interim >> step it would be trivial to write an updateref overlay that supersedes the >> current updateref behavior. > > Updateref processing should imo be kept out of any replication modules, > it is needed in databases where syncrepl isn't configured too. I'm > using it on my subordinate databases (with an artificial updatedn so > that slapd accepts it..) when syncrepl is configured on the superior > glue database without managing all of the subordinate databases. I.e, > the infamous test058... > > I was having the same "chain only on some of the databases" requirement > and is using a patch that allows it. I consider this patch more of a > hack than a correct solution, which is why I haven't contributed it. A > separate updateref overlay apear to me as the best solution. But for > what it's worth, my patch is now in: > > ftp://ftp.openldap.org/incoming/rein-ITS6531.patch > > An advantage with this patch over a new overlay is that it should be > usable without any configuration changes (provided the chain overlay > have already been configured in the databases and found to not work as > expected there that is ... ;-). I didn't quite follow the intended usage for this patch. Are other overlays expected to override the default over_mod_ref() handler? -- -- Howard Chu CTO, Symas Corp. http://www.symas.com Director, Highland Sun http://highlandsun.com/hyc/ Chief Architect, OpenLDAP http://www.openldap.org/project/
moved from Incoming to Software Enhancements
RE 2.5? See also ITS#6942
changed notes
Created attachment 826 [details] rein-ITS6531 patch from FTP server