ets:lookup and ets:update_counter return [] although key/value is there

[Knut Bakke]

Hi,
after upgrading to a more powerfull pentium IV
processor on an ets intensive application we have
experienced that ets:lookup and ets:update_counter
sometimes returns empty list although the key is
there. By doing another lookup/update_couter
immediatly after the failed one, the operation
succeeds (key and value are returned). We see that
this problem only affects a small set of keys (and the
same keys). That is, 2 or 3 keys (and the same ones)
out of about 1000 elements in a ets are having
problems every now and then.

There are parallell ets:update_counter/insert working
on the same keys from other processes, but no delete.

I know that ets should be deadly safe, but it is
strange that the second lookup/update_counter happens
immedialty after the failing one ALWAYS succeeds.

The fault is reproducable only on the machine we see
the problem and for the specific keys.
The test simply lookup a key that we have seen
failing. Doing 10000 ets:lookup's about 4 fails.

We have upgraded to R10B without any improvement.

Any idea?  

My best clue: I know that the memory driver for erlang
has been changed to delete memory areas before
updating them - to reduce memory fragmentation. Could
it be that there is a time slice where the data is not
there?

BR Knut Bakke.

__________________________________________________
Do You Yahoo!?
Tired of spam?  Yahoo! Mail has the best spam protection around 
http://mail.yahoo.com