Mark Thomas wrote:
Remy Maucherat wrote:
Indeed. But do we need to sync the reads in 5.5.x as well or is it
enough just to do the writes? I am confused as you said concurrent reads
were the issue but syncing the writes would fix it.
No, concurrent reads are not the issue. The problem is that when the
HashMap structure is corrupted by concurrent writes (either a put
causing a resize, or a remove), then any read might loop. I believe as
long as the HashMap structure remains consistent, all the next fields
for the entries (which is the problematic field) will have a correct
value at the end of the resize/remove, so an infinite loop on get is
then not possible.
What this issue tells me, also, is that using the default size for the
attributes HashMap may be inappropriate (it is best to avoid resizes,
and the question is how much they happen - if at all, maybe it is remove
which is most often used).
Sorry if I am being dense here ;)
No problem :)
Rémy
---------------------------------------------------------------------
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]
