I'm trying to replicate the issue on a staging website as i agree, it would be preferable to work on a staging copy. I will let you know regarding this.
While i don't want us to make any changes to the live site, at least not yet, you did analyze the error on the live site previously to send the data to tier 2 support right ? Were there any indications regarding the source of the problem ?
Yes, our 2nd Tier Support was able to see the issue(error) on your live site. Unfortunately, they cannot pinpoint the source of the problem yet and would need to perform further troubleshooting on the live site. Please let us know when you're ready to let us check on the issue.
Dear kthorenN and Kor.
we have a similar issue with WPML and Redis object cache. Our English-language (original language of the site) shop page gets stuck now and then with a fatal php error. Error is in allocating memory. We do have enough memory all over, so the error is somewhere else. Error is very difficult to reproduce, but occurs at least once a day. I disabled the Redis object cache and the problem went away.
I would love to get a message once the WPML team solves the problem and I can activate the Redis object cache again!
Thank you for the message Janne, i am glad to know that i am not the only one having this issue. I do hope we can find a solution to this problem so both of us and others can use their object caching if they face this issue as well.
Kor, as i was unable to reproduce the error in any way on a staging copy of the site, the only remaining option is to do tests on the live. Before i accept that, would you be able to give me some idea as to what the potential changes could be, how long my website might have issues while someone is working on it and what the overall risks are depending on the changes you'd be potentially applying or not applying to the overall website ?
As i have a fair amount of trafic, i would like to make sure it wouldn't be a big disturbance to my visitors.
In response to Khoren's inquiry, our Second Tier Support team has requested permission to make modifications to your server files and data. Rest assured, we will make every effort to keep any downtime to a minimum and avoid disrupting or disabling your site. Nevertheless, as a precaution, we strongly recommend that you perform a full backup of your website before proceeding.
Your 2nd Tier Support team has permission to make modifications and hopefully fix the issue related to the object caching.
If there are any changes that are made to WPML settings, files, data, etc. I would appreciate it being noted and shared with me as information for me to know, in case it's needed for my hosting support or for other potential reasons. As some changes may also affect other plugins or functions of the site potentially.
I've notified our 2nd Tier Support and they will run troubleshooting on your production site now. Also, we will keep note of the changes that we've made to fix the issue on your Website/server.
I have feedback from our 2nd Tier Support and this is what they mentioned.
I tried to replicate the issue on the live site again but I can't now. I activated the object cache pro and hit the product URLs for 30 minutes during that time I also set up a script to keep pinging a single product URL in a loop, but found no error.
It seems this was fixed in WPML 4.6.6.
Could you please give this another try on your production site and see if the issue is still there?
I tested on my side and was able to immediately replicate the issue. This time i didn't even have to wait.
Here is what i did : after activating the plugin and enabling object caching (make sure they did go to the plugin and enable the system itself), i immediately Cleared the WP Rocket cache, cleared used CSS and purged CDN, i also flushed the object cache.
After doing this, as mentioned in the previous messages, my main domain urls (from homepage to any other page) stopped working while the pages with secondary languages were still loading.
I hope they will be able to replicate it again and find the solution.
I've tested it just now and I'm getting the "500 Internal Server Error" after activating the Object Cache plugin. I've shared this feedback back to our 2nd Tier Support for a second look.
Very well, i was just getting on the website and noticed the error and disabled the plugin thinking i had left it activated during my own test an hour ago. Then only thought you might be already doing some tests. Thanks for the confirmation.
Thank you very much for your cooperation. We have partially found the issue.
WPML has a function that translates the terms ID to the current language. When the Object Cache pro plugin is active this function goes into the recursive calls and takes all of the PHP memory until exhausted.
This is not a known issue and is the first time we have seen this. Technically it is not possible by the function design we have a condition to prevent this but it is failing due to bad cached value.
We logged some call stack and found product category ID 20 translated to 224 (Spanish ID) on the English page and due to this bad value condition is failing.
However, it is still not known how these bad values are cached. I tried to replicate this with the copy of your site with no success and also on a clean install but no success with the Redis caching.
Since we have limited capacity to debug this on the live site we can't run step debugging. This is not possible to exactly find out the problem.
But I will report this to our development team to see if we can patch the code based on assumption.
Right now, I found if I exclude the caching of "terms" group from persistence caching, I can not replicate the problem anymore.
So I have added this config in WP_REDIS_CONFIG hidden link
Hi Sumit, thank you for looking into this issue and the provided information.
What you are saying confirms the thoughts the hosting support had regarding some problem creating an abnormal amount of constant server calls until the exhausted PHP memory throws an error no matter how high we'd put the limit.
Am i correct in understanding that this seems to be related only to that one product category or is it just an example that you were able to see in the call stack you logged to review and there are probably others ? If it is one product category, maybe deleting and recreating it so it has new IDs could solve the problem ?
It is a bit weird that we are only able to replicate the issue on the live site, there is something specific related to the live site that creates the conditions for the conflict to happen. Maybe it came through the normal process of using the plugin and working on the website, making some changes or something that created a conflict through a lingering piece of data or code or something like that... Something that doesn't come when it's copied.
But it is also important to note that there was another WPML user that answered to this thread and mentioned having the same issue. So it is not a unique case, though it does seem to be rare and probably specific to a combination of various factors that have to be similar to make this happen.
Could you please explain what excluding the caching of "terms" group actually entails ? I understand the concept of excluding some groups from the persistent cache to avoid problems. But what does the "terms" group include ? Does it mean that the product category pages content won't be cached ? Or all pages that have ID ? Or ? Sorry if i'm missing something obvious.
Can this have any potential impact on any aspect of the website usage both for admins and normal visitors ? Any impact on SEO related configuration ? Were there any other changes to the options in the website i should maybe be aware of just in case ?
I'd like to confirm that i enabled object caching and tried to replicate the issue and so far i don't see the error. I will keep it active for now and see if it keeps working.
Despite it working, i would really appreciate your dev team looking into this for a potential true fix of the problem. I assume they'll look into my website and i should keep the accesses open ? Will they investigate this soon or would it take longer ?
PS : I'll keep those plugins while the ticket is open.
Sorry for the multiple questions and thanks for your attention,
Khoren
