Gianni Ceccarelli wrote:
> On 2013-08-31 Kevin Grittner wrote:
>> [Locks without PIDs] are predicate locks related to a
>> transaction which has been PREPARED (for two-phase commit) or
>> committed, but which may still be relevant because there are
>> overlapping read-write transactions which ar
On 2013-08-31 Kevin Grittner wrote:
> Of course, the subject line gives me some pause -- I'm aware of many
> uses of SSI in non-trivial production environments, including
> multi-terrabyte databases with thousands of concurrent users. In
> some cases the performance hit compared to REPEATABLE READ
Kevin Grittner wrote:
> - We don't distinguish between heap relation locks which need to
> prohibit inserts (those caused by a table scan) and heap relation
> locks which don't conflict with inserts (those caused by promotion
> from finer granularity). We would reduce false positives if we
> d
Gianni Ceccarelli wrote:
> At work we have a program that seems to be stressing the SSI
> implementation, and I thought that it could provide useful
> insights to better tune it. In particular, there are a few parts
> that are described as "chosen entirely arbitrarily (and without
> benchmarking)
Hello.
At work we have a program that seems to be stressing the SSI
implementation, and I thought that it could provide useful insights to
better tune it. In particular, there are a few parts that are
described as "chosen entirely arbitrarily (and without benchmarking)",
and we may provide some of
