Yup, totally agree. Will also try to make these more clear in the doc.
We want to queue up the work to "log" and do it in the background. If
that queue gets too full, we drop it right off that cliff :)
Thanks, Samarth!
On 3/2/18 5:47 PM, Samarth Jain wrote:
A couple more points which I
A couple more points which I think you alluded to, Josh, but I would still
like to call out:
1) Writing of these query logs to a phoenix table should be best effort
i.e. a query definitely shouldn't fail because we encountered an issue
while writing its log
2) Writing of query logs should happen
Thanks Nick and Andrew! These are great points.
* A TTL out of the box is a must. That's such a good suggestion
* Sensitivity of data being stored is also a tricky-serious issue to
consider. We'll want to lock the table down and be able to state very
clearly what data may show up in it.
* I
My gut reaction would be to avoid storing queries over the system tables
as they would have more noise than value, but I think this is something
that could be entertained!
On 3/2/18 8:02 AM, Artem Ervits wrote:
+1
Is idea here to collect information on all types of queries even against
Agree with Nick's points but let me augment with an additional suggestion:
Tunable/configurable threshold for sampling. In many cases it's sufficient
to sample e.g. 1% of queries to get sufficient coverage and this would
prune 99% of actual load from the query log.
Also let me underline that
I'm a big fan of this idea. There was a brief discussion on the topic over
on PHOENIX-2715.
My first concern is that the collected information is huge -- easily far
larger than the user data for a busy cluster. For instance, a couple 10's
of GB stored user data, guideposts set to default 100mb,
+1
Is idea here to collect information on all types of queries even against
system tables? It would be nice to keep counts of how many times a query
was executed over time. In my mssql days we also had ability to check when
was the last time update stats ran and all kinds of impact information
Hi Josh,
Thanks for starting this discussion. Overall the design looks good to me!
Actually I have an internal implementation for logging user queries to text
file over PQS. Let me explain our use cases. We wanted to find out
following facts using our ugly hacks:
- Running queries against PQS
-
Any feedback from folks? Not sure if the silence should be interpreted
as ambivalence or plain old being busy :)
On Mon, Feb 26, 2018 at 4:57 PM, Josh Elser wrote:
> Hiya,
>
> I wanted to share this little design doc with you about some feature work
> we've been thinking
Hiya,
I wanted to share this little design doc with you about some feature
work we've been thinking about. The following is a Google doc in which
anyone should be allowed to comment. Feel free to comment there, or here
on the thread.
https://s.apache.org/phoenix-query-log
The high-level
10 matches
Mail list logo