memcached seems unable to return values on "get" but can "delete"

[Gabe Schine]

wow. so, problem solved. essentially, our cache server's clock has  
been drifting (for a while apparently) and it's about 7 hours ahead  
of the server on which we compute our expiry timestamp. so, the keys  
that worked were always ones with infinite expiry times.

sorry to bother you all. :)

-gabe


On Feb 2, 2007, at 2:17 PM, Jehiah Czebotar wrote:

> i don't have any ideas yet; but it might be good if you can mention
> which memcached server version you are using, cmd line options to
> start memcached, and os rev etc, and if you can write and post a
> simple script that duplicates this problem that's always useful to
> send to the list.
>
> --
> Jehiah
>
> On 2/2/07, Gabe Schine <gabe.schine at gmail.com> wrote:
>> hey there,
>>
>> so we have a single memcached installation serving about five
>> servers. it has been working great, however:
>>
>> certain types of data, usually blobs of about 6k in size, seem to get
>> "lost" or something. we store the value with 30 min timeout and a
>> flags value of "8", and subsequent gets just return "END", no value.
>> HOWEVER, what's odd is, I can contact the memcached server and tell
>> it to "delete" that key, and that succeeds, leading subsequent
>> deletes to return "NOT_FOUND", so i'm thinking memcached finds the
>> data only when i want to delete it.
>>
>> other data, usually ~600 bytes, seems to work fine. there are two
>> differences between data A (doesn't work) and B (works):
>>
>> 1) the keys for A are formatted like this:
>>
>>                 Mind.1:-73828371281
>>
>>         whereas for B they are something like:
>>                 xxx:1234567890ABCD
>>
>> and 2) data B has an infinite timeout (value of 0) instead of the
>> usual 30 mins
>>
>> lastly, if i try to "set" a value using the perl memcached client, i
>> always get back "SERVER_ERROR: Out of memory" even though i KNOW the
>> server isn't out of memory, and our Java production install client
>> seems to be able to set keys just fine (barring the above anomaly).
>> i'm at the end of my rope here and am just scraping the floor for
>> ideas -- anything at all would be hugely appreciated!
>>
>> thanks very much,
>>
>> gabe
>>