 |
History
| Log In
|
|
|
|
im not sure about youre idea. some things will work
different: concurrent use of servlet-context, only one set of war-settings (web.xml) for multiple portal-instances. what if you want to use different filters. for instance: portal-a uses cas(-filter) and portal-b dont. we pack some additional resources to our portal-web-xxx.war. each portal-instance can have different additional resources. we would have to be more careful with this things ... use some naming-convention ... hmmm ... may be your plan are ok though ... handle with care ;-) This is a great idea. I've run into the same exact issue. At the moment, I'm just working around the issue since I'm providing mostly community functionality to users. That said, my intentions were to do exactly what you are doing (when I got time). It seams strange that this wasn't already done. I can see a great deal of ASP applications that could benefit from your changes. A HUGE benefit with your changes I see is system resources. I haven't done any tests to see how much each portal "costs" but I do believe it adds up very quickly.
As to Helmut's comment, I understand what he is getting at but I think is a situation where it depends on the application. In the ASP model I use, each customer would have similar functionality and just need their data properly scoped (which is what we are using the company id to do). I'll try your code out this week and see what happens. Any word from seasoned developers? Hi Atul,
I'm planning to take a look at the patch later this week or early next week. One of the first things I want to do is to see if the current functionality used by some, as Helmut, is kept. If you can download and apply the patch, test it and help fixing any issues that you find that would be helpful. Well, as Helmut says (and, I think, I also pointed out in some comment), there would be some hidden places where a deeper look would be recommended :)
Of course, as long as this modifications result in serving N portal instances with just one webapp instance, the main implication is that all those N portal instances are sharing the same war settings. That would be a problem in some situations.. but.. if you need different instances to have "personalized" war settings you always can fall back to the "old" or "original" ASP model (1 instance <-> 1 webapp) or group instances in sets that share the same war settings and deploy those instances in separated weapps. Of course, a more sofisticated solution would be ok, but I think that solution diverges from my main objective when developing the modifications. I was thinking about sharing a webapp among different portal instances that *actually* didn't need war specific configurations (that is: they share datasources and all other stuff defined in the war configuration -being filters one of these settings?). This would result in the possibility of ASP deployments where the only possible per-company configurations are those available in portal.properties, by means of portal-companyId.properties files (and this actually works, as far as I've tested). I've not taken a detailed look of the possibilities available by means of war settings configuration, but I guess that it would not be so easy to "move" to application level, as long as they are strongly related to a "classic" weapp concept, and what I was proposing, actually, was to have N portals (companyId instances) as 1 webapp. As you see... 1 webapp -> 1 war settings (in the originaly idea). Anyway, I think this would be an interesting discussion. I'll take a look to it but I think that the core developers would be able to expose arguments much more founded and stronger than mine (basically, by knowledge of the software). And remember... anybody needing war-specific settings can always fall back to the 1 portal <-> 1 webapp model :) There's also another issue that Helmut points out in it's comment, and it's the concurrent use of servletContext. To be honest, I didn't realise that... I was making an "optimistic approach". I'll take a deeper look at it, and I hope concurrent access would be OK... actually, concurrent access to it should be already expected and well managed, as long as it is shared by different requests. I'm assuming you are afraid about some not "companyId-alized" behaviours around it? I'll be happy to help out. I've been a bit swamped and haven't had a chance to try out the patch yet. Hopefully will get to it today.
--atul This is really great feature. I hope this makes into LEP 4.3 even if its in just experimental stage. Currently we are using virtualhosted communities but there the problem is that you can't limit the visibility of other communitys users to other communities.
Hi all.
So I've been working on this for the last couple of days. Here is a summary of where I am so far... I've created a VERY basic and PRELIMINARYY portlet addition to enterprise admin that will add and search for "subcompanies". This basically creates an a company and sets it associated account's parentaccount to the accountid of the current logged in company/site. It then uses checkCompany to create the default records. I then create a community in the new company with the same name as the company. It still needs layouts. I've modified VirtualHostFilter to convert the virtualhost into a companyId that is set as an attribute on the request object. My next step is to convert the remaining references to the company object to come from the request instead of from the portal context and then test... There is still a lot to be done regarding the initialization routines, and validating the change. I'm new to portlets/struts and was wondering if anyone would like to collaborate... I think the new accounts tab under enterprise admin should be modified to include all the subtabs that are avaiable below settings in the admin portlet. I'll let you all know how it goes.. Thanks guys for the great application and for all the work toward converting the IDs to longs just remembered. A cool (IMHO) connection I made is that when the community for the new company is created, its layoutSet's virtualHost is automatically populated based on the portal url for the company -- perhaps others have made this connection -- any thoughts on why this shouldn't be done. Hi Atul,
Thanks for your work on this. I have a question, is your work based on the patch provided by Jose Oliver? Based on your description it seemed to me that you are following a different approach. So I had written an epic comment, but then clicked attach multiple files and lost all of it... aargh. I've attached 2 subversion patches, and my struts and tiles def files.. The patches are based on Rev 6202. It needs LOTS of TESTING, but it does work in adding child companies, and rendering them based on community virtual hosting. When adding a company I used the portaURL to determine a virtualhost name. Here are the assumptions I made to get this to work... 1. It is safe to attach the companyID to the request attribute during the virtualhostfilter 2. It is safe to iterate across the initializations within mainservlet for each company 3. It is safe to get companyid from the requestAttribute and use it to define the company context in all cases 4. It is safe to determine the companyid based on the layoutset's virtual host matching the request's hostname (i guess this is the really important assumption) I also created a very very preliminary tab in the enterprise admin portlet.. It works but gives an error message at the end.. Can someone help with creating a better portlet / copying the Admin/Settings subtabs to the new Accounts tab in EnterpriseAdmin. We probably need to also add a new permission called Add Company.. The webid needs to be of the .com/.net, etc variety because the admin email address is created based on this. The additional tab allows additon of new companies. My comments can be found by searching for atul in the code I forgot what else I had originally wrote, but if you have questions, please email me... or skype at atul.harbor9. I'd love to collaborate with someone to make this production ready. Thoughts.... Feedback please... Hi Atul,
Thanks for your patches. I (or someone else from Liferay) ll have a look at it, although it'll have to wait until 4.3 is released because this is a big change that would have the effect of delaying the release a lot. I hope that's ok with you. Also it would be great if you and Jose could coordinate/integrate both approaches. On 5/9/07, Atul Patel <atul@harbor9.com> wrote:
I had compared your code to the existing code to get an idea of what you were doing, but the actual code itself is all fresh... Yes, I've seen the patch and lots of code there! :) In my scenario, I believe in zero filesystem access to run services. That's my ideal scenario also, but since LEP uses a bunch of properties from system/portal.properties, I didn't go that way. I agree with you in the sense that, as long as it seems, everything should be able to be managed from UI Everything should be provisionable and controllable from the interface. That said, a few UI portlets will need to be made I think this would be basically: extending admin portlet (as you suggested) to allow omniadmin/admins/people-with-permission to create/manage/remove companyIds and a "little" extension to allow to manage .properties for each company by using the UI. These are modifications in "UI land only". (something I need to find help on.) My scenario basically has my master company creating children companies with the potential that those children companies may create their own (but that is way in the Seems ok future and so need a permission system to allow me to see the add children company and not other admins.) That's ok also, permissions should be there I believe (though I didn't test it) that the changes also allow for multiple virtual hosts within any single company. I think they should be compatible. In my approach, I'm able to associate *.mydomain.tld to a companyId, so I don't see why it shouldn't work with domain1.mydomain.tld,domain2.mydomain.tld, etc. I use the "virtual hosting" mechanism to essentially accomplish the same thing as I believe you are trying to do without additional overhead of a secondary properties file to manage. This allows the I think we're also pursuing the same objective (sharing a webapp among different instances). For me, having a secondary property file was not a big overhead, since companyIds are not created/destroyed every second :) Anyway, having everything available from the UI could be great. system to be very dynamic... It would be nice to add a properties portlet that would configure the properties for each company. I think that's one of the key points. Moving companyIds management to top level (UI) seems reasonable only as long as you also provide tools to manage *everything* and I think the only thing missing there are those in portal-XXX.properties. Because of the iteration on multiple companies during the init, I believe a performance issue may occur once hosting many sites... I haven't run any metrics to determine the "cost" of each additional company. I have run very simplistic tests, and a 60-instances startup with my modified version has no noticeable (maybe 5-10 seconds, a total of around 40 seconds) overhead. Having this 60-instances with the native approach (separated weabpps) would have been... don't know... maybe 30/45 minutes? **minimum** I wonder if we are both trying to achieve the same goals... perhaps Maybe. As I've told, I want to be able to have *one* webapp serving *many* companyIds. I'm starting up a little bussines based on Liferay ASP hosting (offer Portal [company] instances to end users, without worrying about installation, upgrades, backups, etc.), so I want to maximize resources usage. Basically, this is a next-level service compared with RimuHosting Liferay Hosting (I think). What are your goals? we should talk and find a way to divide the work and share the results... :) It would be great... I'm finishing the setup of the site, so I'll launch the service using my approach, in order to have it up and running asap, but afterwards I'll able to collaborate with you and the Liferay guys in order to integrate the ASP model support inside liferay. Thanks for the fast response, I can't wait to hear the good and the bad (please) about the changes... don't worry about hurting my feelings, I just want to improve the product. Ok, we'll continue talking. I've seen Jorge has answered but I can't read the response now (going late to a meeting :) ), I'll continue later. Best, Jose Do you think we shoud start to write down a little design doc / concept doc / whatever, in order to list what we think LEP needs to support pure ASP and how to support it? Maybe we can use the "liferaypedia"? don't know... I think that some involvement from someone in the Liferay core team would be great, since there are som architectural issues that need to be reviewed by someone with a global scope
Hi Jose,
That's a very good idea, I've created a new page in the Proposed Projects section of LiferayPedia about this. You can find it here: http://wiki.liferay.com/index.php/Dynamic_Management_of_Companies I'll keep and eye on it and will other Liferay developers know about it once it get's more content. Since this is a feature I need I ported Jose's original patch against latest trunk. I'm hosting multiple sites with their own user base on a single server with limited memory. I did some refactoring since there was some duplicate code in Mainservlet and LuceneServlet. I hope we can coordinate our efforts through the wiki to get a solution that is good for everyone. In the mean time try this if you want it in LEP 4.3.
Mika, nice job.
As you point out, the modifications I submitted needed some refactoring (noticeabilly most copy/past code in MainServlet and others :) ), since they started much as a proof-of-concept than any other thing. I'm busy these days so I've not had time to get into the Wiki, but that's something I'm willing to do in the next days. And with more hope, since I see that there are some more people willing to have this feature available in next releases. How is going your experience hosting those multiple sites? Any "hidden traps" in the multicommunity patch? Wow. This is pretty nice stuff.
Mika, can you coordinate some "trunk" patches? Perhaps we can get this into 4.3, and if not, then in 5.0 for sure (the next release after 4.3). Working on merging Mika's 5/10 patch. It's actually very lean. The bulk of it is in MainServlet, VirtualHostUtil, and in LuceneServlet. Will give an update when it's completed.
Hey Mika, I'm having some merge issues. Can you send me your MainServlet.java and your LuceneServlet.java? Thanks.
Hi Brian. What do you think about having company_ids come from the database instead of from properties files? I'd be happy to spend some time coding for that. (Its much easier for me to replicate just a database instead of also replicating configuration files ) for backup/redundancy purposes. -Atul t's very nice that moving on something positive regarding this patch. I also need it badly .
Actually i am testing this also - to be exact, Mika's patch. I'm testing on tomcat/winxp for now, but in few days i'm going to upgrade tomcat/LP.4.2.2 on my Linux server to LP.4.3 trunk build. For now almost everything seems to be working fine on functionality, except few issues i found in my logs. More details on errors see in my post on forums thread http://www.liferay.com/web/guest/devzone/forums/message_boards/message/29234 Really it's killing feature, and i hope so much that it will be included in 4.3. Even if still in unadvanced stage, better it would be there, if it is not impacting any other functionality. IMHO very first thing to do it would be recheck twice about any possible issues, and don't go into advancing features to much for now, otherwise it can stretch till 5.0... Please don't do that... About storing mappings in db , it would be great, but... not exactly now :). Current functionality is quite enough for normal work, anyway there is lots of config files which u have to backup anyway, and with storing mappings in db we will need also webUI for editing, then something else, then something more... and we will have something working without patches maybe in 6.0 :)... Please lets make first inside trunk what is working now , and improvements can be done later also. Like Atul can do already some codding nd testing, even for that can be opened one more issue or sub issue for this issue, but please don't mix it up here everything at once and delay all this nice thing till next year... Lots of coding needs lots of testing and i want things working :) What others think on this ? i think there was more ppl who needs it working ? :) Brian here's my MainServlet and LuceneServlet
Here is my original patch updated against latest revision (6391) in trunk
Refactored my latest patch a bit. Instead of using a thread to poll the config file I moved it to use JobScheduler.
Ultimately I think we should use database to hold the config but that is something that should be done for 5.0. I'd rather see 4.3 with this properties file based confg than delay 4.3 release. Nice to see all this movement around this issue!!
I agree with Mika and Andrius regarding the db usage for dns->companyId mapping. I think it would be great to have the main functionality in the 4.3 and that the db usage for that mappings could waint until 5.0. Actually, there are some other issues that need to be addressed in order to have a full "GUI-managed" ASP setup. Actually, despited all the mappings were managed using the GUI and stored/retrieved from DB, many other configuration setup would still require command line access in order to customize instances. For example: mailhook to use ("real" or "dummy"), languages for each portal, etc. (I have a list that I use for each host I include in my asp services, so maybe I can review it when I have access to my hosts and include it here). All this is related with the possibility of "moving" some .properties configurations onto GUI-space for full-GUI-manageable companies. After having the main functionality (multiple companyIds, mappings obtained from properties file) inside the core, I think we can take some time to discuss the best strategy to get all this manageable from the admin UI (look at the wiki page, I think there are many thinks to take in consideration), make it fully-compatible with VirtuaHostingThroughCommunities, etc. See
Props goes out to everyone who helped out with this issue. Go to the Admin portlet -> Go to Instances You can add new companies via the GUI now. This has been tested with Community Virtual Hosting AND enterprise/company/portal instance virtual hosting. You can now add companies and communities on the fly. Had to make multiple commits because of SF.net SVN choking when it's too many files.
See revision 6408 to 6416. See
|
||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
 |
|
Improvement
Closed
Minor
Ability to serve multiple companies/portals with just one instance of Liferay (only one webapp)
I hope it will serve at least as proof-of-concept
Main modifications:
* Modified some servlets to ignore static value of companyId and use a companyId based on the host name requested
* Added class to maintain mappings from hostname/domain to company id
* Modification on some other -non servlet- classes that also made use of a static companyId or that just take in care one companyId (for example, extending Lucene services to have one thread per company)
* Modifications to some external systems interfaces (ex. mailadmin.ksh, made it "asp friendly" and adding a new AspShellHook that configures environment variables before invoking the shell hook to pass the company id to the script)
* Some minor bug fixes (AspShellHook -> mailadmin.ksh interaction didn't work ok if user has spaces in the second name, inconsistences between userIds created by mailadmin.ksh and the ones used by the mail portlet, etc.)