Is it just me who keep getting some sort of strange error on searches on :en:?
The symptoms so far include showing the Google/Yahoo boxes and an odd error message indicating some sort of bad return from an IP in the [10.*.*.*] range. My IP knowledge is rusty: is that a private range?
Sorry I can't be more precise: the error sometimes goes away when I try again, and did this time :-(
Phil Boswell wrote:
Is it just me who keep getting some sort of strange error on searches on :en:?
The symptoms so far include showing the Google/Yahoo boxes and an odd error message indicating some sort of bad return from an IP in the [10.*.*.*] range. My IP knowledge is rusty: is that a private range?
Sorry I can't be more precise: the error sometimes goes away when I try again, and did this time :-(
At the moment, we have three search servers: maurus (10.0.0.16), vincent (10.0.0.17) and coronelli (10.0.0.230). For reasons not quite understood, they sometimes hang, accepting connections but closing them immediately. Yesterday, Brion set up an hourly restart of the daemons to mitigate this. Most of the search errors before then were probably due to this problem. The server you get for any given query is random, that's why it often goes away when you try again.
The other potential problem is a timeout: if the search server takes more than 3 seconds to respond, then an error will be returned. This could happen due to high load, or due to random temporary events. Looking at the profiling data I gathered yesterday, it's likely that this has been happening a fair bit during peak time.
-- Tim Starling
The symptoms so far include showing the Google/Yahoo boxes and an odd error message indicating some sort of bad return from an IP in the [10.*.*.*] range. My IP knowledge is rusty: is that a private range?
At the moment, we have three search servers: maurus (10.0.0.16), vincent (10.0.0.17) and coronelli (10.0.0.230).
Do we have to expose this detail to non-technical users? The above user clearly knew a fair bit about IPs, and the message was still meaningless to them; imagine what an average non-technical user might think.
Timwi
Timwi wrote:
The symptoms so far include showing the Google/Yahoo boxes and an odd error message indicating some sort of bad return from an IP in the [10.*.*.*] range. My IP knowledge is rusty: is that a private range?
At the moment, we have three search servers: maurus (10.0.0.16), vincent (10.0.0.17) and coronelli (10.0.0.230).
Do we have to expose this detail to non-technical users? The above user clearly knew a fair bit about IPs, and the message was still meaningless to them; imagine what an average non-technical user might think.
We have detailed error messages so that sysadmins can debug problems when users report them.
-- Tim Starling
Tim Starling wrote:
Timwi wrote:
Do we have to expose this detail to non-technical users? The above user clearly knew a fair bit about IPs, and the message was still meaningless to them; imagine what an average non-technical user might think.
We have detailed error messages so that sysadmins can debug problems when users report them.
I've switched it to automatically cycle to the next server if the first one doesn't respond; if all three fail it'll now print out:
"There was a problem with the wiki search. This is probably temporary; try again in a few moments, or you can search the wiki through an external search service:" [followed by the google and yahoo search forms]
(This is localizable and configurable as MediaWiki:Lucenefallback)
-- brion vibber (brion @ pobox.com)
"Timwi" timwi@gmx.net wrote in message news:dkq22f$e7q$2@sea.gmane.org... [me originally, then I think Tim Starling]
The symptoms so far include showing the Google/Yahoo boxes and an odd error message indicating some sort of bad return from an IP in the [10.*.*.*] range. My IP knowledge is rusty: is that a private range?
At the moment, we have three search servers: maurus (10.0.0.16), vincent (10.0.0.17) and coronelli (10.0.0.230).
I managed to capture one: it was "vincent" * Internal error: no valid response from search server (10.0.0.17)
Do we have to expose this detail to non-technical users? The above user clearly knew a fair bit about IPs, and the message was still meaningless to them; imagine what an average non-technical user might think.
You flatter me, for which thank you, but you're reasonably correct. If we're going to expose messages like that to naive users, they should be bracketed with a warning, and possibly some idea of someone to inform of the problem.
wikitech-l@lists.wikimedia.org