On Tue, Aug 15, 2017 at 5:23 PM Sean Busbey <[email protected]> wrote:
> Say we replaced it with brave for 2.0, Phoenix tracing through to hdfs > would be just as broken as if we pulled it out entirely, no? Yep! From the perspective of Phoenix, HTrace API is our public API. We should consider this as we do other places where 3rd party deps leak into our API. Just as we removed Guava/Service, I suggest we should remove HTrace/Span. If we want to support tracing, we should provide our own API for it, and provide connective tissue for interop with whatever 3rd party libs our users want to use. Or at least interfaces that downstreamers can use to implement that connective tissue. HBase Spans -> HTrace Spans would be a good demonstration of such an API, and give us what we need to restore Span continuation through HDFS. All this requires an owner, for which tracing in HBase has none. Maybe our Tamas has some vested interest in being that owner, having brought up the topic to start with? On Aug 15, 2017 7:18 PM, "Nick Dimiduk" <[email protected]> wrote: > > > Agreed that ripping with no replacement would be too extreme and short > > notice for 2.0. > > > > On Tue, Aug 15, 2017 at 5:14 PM Andrew Purtell <[email protected]> > > wrote: > > > > > > If Phoenix devs are advertising this feature to their users, > > > > their contributions re: HBase and HDFS maintenance would be well > > > received. > > > > > > The counter point to this more mindfulness on the part of HBase devs to > > > interface compatibility impacts would go a long way. See HBASE-18431 > > > and PHOENIX-4076. > > > > > > Ripping out entirely the only tracing interfaces available is a more > > > extreme example, although not without cause. (The state of HTrace.) > > > > > > > > > On Tue, Aug 15, 2017 at 5:10 PM, Nick Dimiduk <[email protected]> > > wrote: > > > > > > > IMHO, we should be actively pruning half-implemented, testless, > > ownerless > > > > code. HTrace queues up well for that chopping block. Long back I > wanted > > > to > > > > see this particular feature though, but I cannot volunteer for the > role > > > > currently. If Phoenix devs are advertising this feature to their > users, > > > > their contributions re: HBase and HDFS maintenance would be well > > > received. > > > > > > > > I question the choice of HTrace as the tool for the job, given the > > > relative > > > > lack of activity in that project vs. the myriad alternatives out > there. > > > > HTrace has been incubating for 2.5 years, has seen only 5 releases, > > with > > > no > > > > release in the last 18 months. I haven't evaluated the field in a > > while, > > > > but there's probably something with much more mature tooling and > > support. > > > > For instance, openzipkin/brave has much more activity more recently. > > > > > > > > -n > > > > > > > > On Tue, Aug 15, 2017 at 2:28 PM Stack <[email protected]> wrote: > > > > > > > > > On Tue, Aug 15, 2017 at 12:38 PM, Andrew Purtell < > > [email protected]> > > > > > wrote: > > > > > > > > > > > Doc update to indicate HTrace is unfinished is fine. > > > > > > > > > > > > > > > > > We could do this. Update jar and doc it as broke. > > > > > > > > > > > > > > > > > > > > > Phoenix uses HTrace to support query tracing. It "works". > > > > > > > > > > > > > > > Nit: I didn't think this was htrace. It was an old version pulled > > into > > > > > phoenix world where it lives storing to phoenix table reporting on > > > > > phoenix-only spans. > > > > > > > > > > > > > > > > > > > > > A colleague made > > > > > > a contribution to HDFS to fix a case where spans could get > muddied > > > > around > > > > > > the WAL. We are carrying that patch on top of 2.7 for the moment > > but > > > > IIRC > > > > > > it's in 2.8. > > > > > > > > > > > > Removing all of our HBase level spans doesn't make sense to me > when > > > > spans > > > > > > at HDFS level will not be removed, nor at the Phoenix level. > > > > > > > > > > > > > > > The spans in hbase are just hanging until they get the htrace-4 > > revamp > > > > and > > > > > the dots get connected. > > > > > > > > > > > > > > > > > > > > > We hollow out > > > > > > the middle. I think this is worse than leaving things at a half > > > > finished > > > > > > state. At least in the half finished state it might get finished. > > > > > > > > > > > > > > > > > Nod. > > > > > > > > > > > > > > > > > > > > > What are the most pressing issues? I'm up for supporting an > > "HTrace" > > > > get > > > > > > well. I'll need your (community) help for getting patches into > > Hadoop > > > > if > > > > > > need be. > > > > > > > > > > > > > > > > > HTrace needs care and a viewer (zipkin I think now it has come back > > > > alive). > > > > > > > > > > HBase needs a couple of narratives wired through. > > > > > > > > > > Ditto HDFS. > > > > > > > > > > St.Ack > > > > > > > > > > > > > > > > > > > > > > > > > > > On Tue, Aug 15, 2017 at 12:10 PM, Stack <[email protected]> > wrote: > > > > > > > > > > > > > On Tue, Aug 15, 2017 at 11:22 AM, Andrew Purtell < > > > > [email protected]> > > > > > > > wrote: > > > > > > > > > > > > > > > -1 to removing HTrace from HBase. > > > > > > > > > > > > > > > > Upgrade to the desired version, that would be fine. > > > > > > > > > > > > > > > > > > > > > > > âtâ > > > > > > > > > > > > > > What you thinking Andrew? > > > > > > > > > > > > > > Trace is broke whether we upgrade or not. As is, we mislead. We > > > give > > > > > off > > > > > > > the impression that hooking up hbase and hdfs via trace tells a > > > story > > > > > > about > > > > > > > request traversals when it at best you'd get a wonky view. > > > > > > > > > > > > > > HTrace needs work. HBase and HDFS also need work before we can > > > > suggest > > > > > to > > > > > > > our users this system is viable debugging. > > > > > > > > > > > > > > What you reckon Andrew? > > > > > > > Thanks, > > > > > > > S > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > On Tue, Aug 15, 2017 at 5:54 AM, Tamas Penzes < > > > [email protected] > > > > > > > > > > > > wrote: > > > > > > > > > > > > > > > > > Hi All, > > > > > > > > > > > > > > > > > > As far as I know the actual integration of HTrace 3.2 into > > > HBase > > > > is > > > > > > far > > > > > > > > > from satisfying. > > > > > > > > > > > > > > > > > > This version of the library is old, buggy, not really > > > maintained > > > > > and > > > > > > > the > > > > > > > > > migration to version 4.x is quite hard ( > > > > > > > > > https://issues.apache.org/jira/browse/HBASE-14451) and > looks > > > > like > > > > > > the > > > > > > > > task > > > > > > > > > is not getting much closer to the goal. > > > > > > > > > > > > > > > > > > This is why I think we should remove HTrace now and if we > > still > > > > > want > > > > > > to > > > > > > > > > have some art of tracing (I think it would be useful) then > > > start > > > > it > > > > > > > from > > > > > > > > > the baseline. > > > > > > > > > I have created a ticket for this task: > > > > > > > > > https://issues.apache.org/jira/browse/HBASE-18601 > > > > > > > > > The removal could be quite fast and reach version 2.0.0 > too. > > > > > > > > > > > > > > > > > > What do you all think? > > > > > > > > > > > > > > > > > > Thanks, Tamaas > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > -- > > > > > > > > Best regards, > > > > > > > > Andrew > > > > > > > > > > > > > > > > Words like orphans lost among the crosstalk, meaning torn > from > > > > > truth's > > > > > > > > decrepit hands > > > > > > > > - A23, Crosstalk > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > -- > > > > > > Best regards, > > > > > > Andrew > > > > > > > > > > > > Words like orphans lost among the crosstalk, meaning torn from > > > truth's > > > > > > decrepit hands > > > > > > - A23, Crosstalk > > > > > > > > > > > > > > > > > > > > > > > > > > > -- > > > Best regards, > > > Andrew > > > > > > Words like orphans lost among the crosstalk, meaning torn from truth's > > > decrepit hands > > > - A23, Crosstalk > > > > > >
