zeo clients not synchronized, modifications not shown
Hi there, I have in production a Zope server from many years now: - Zope 2.11.4-final, python 2.4.6 - 2.4.6 - ZEO with 3 clients via sockets on the same machine - running a custom CMS (not Plone based) Last weekend, without touching the server or the code, a strange problem starting to occur and is repeating since: - when I try to modify an object (e.g. only the title) the saved data is not available in the browser after that. If I look in the ZMI the transaction is there. If I refresh the page the modification appears - also happens on an add or a delete (the object is deleted bu still appears in the listing) My impression is that the save itself is handled by lets say client1 and then the redirect and render of the edit page by a different one - client2. I tried different things: 1. pack data fs and restart all clients and zeo 2. if i try to change the title of the object via ZMI Properties tab it works fine - I always see the changes. On our CMS edit page we do an explicit recatalog of the object after the save. I thought maybe the catalog is broken somehow so I updated it - problem still there. 3. I checked the data fs with fstest.py, also checked for broken BTreeFolders - no issues detected. 4. I put the production database on a development server without zeo, just a single zope instance - everything works just fine. So, I can reproduce the problem but now clue so far about the cause. Any idea on how to handle and fix this? Anything will be highly appreciated. Thank you, Dragos -- Dragos Chirila objectValues@gmail.com (+4) 0722 395375
On 11/5/15 12:29 PM, Dragos Chirila wrote:
My impression is that the save itself is handled by lets say client1 and then the redirect and render of the edit page by a different one - client2. I don't know what is going on, but let me ask a few obvious questions. How do you assign web requests to zeo clients. If the redirect is handled by a different zeo client, then maybe it has not gotten the cache invalidation. Maybe in your web page display, you can display the process id, to verify if indeed it is a different zeo client which is the source of the problem. At that point, put some logging into the cache invalidations to see if they are happening correctly, and before the new page is redisplayed. Of do something on your load balancer to make sure that each ip is served by the same zeo client.
And why did the problem suddenly start happening? Wierd. Hope that helps.
Hi Christopher, Great pointers, thank you very much for your response! I use pound for load balancing (from the beginning - nothing changed here). I will investigate using your indications. Thanks again, Dragos On Thu, Nov 5, 2015 at 2:03 PM, Christopher Lozinski < lozinski@freerecruiting.com> wrote:
On 11/5/15 12:29 PM, Dragos Chirila wrote:
My impression is that the save itself is handled by lets say client1 and then the redirect and render of the edit page by a different one - client2. I don't know what is going on, but let me ask a few obvious questions. How do you assign web requests to zeo clients. If the redirect is handled by a different zeo client, then maybe it has not gotten the cache invalidation. Maybe in your web page display, you can display the process id, to verify if indeed it is a different zeo client which is the source of the problem. At that point, put some logging into the cache invalidations to see if they are happening correctly, and before the new page is redisplayed. Of do something on your load balancer to make sure that each ip is served by the same zeo client.
And why did the problem suddenly start happening? Wierd.
Hope that helps.
-- Dragos Chirila objectValues@gmail.com (+4) 0722 395375
participants (2)
-
Christopher Lozinski -
Dragos Chirila