Re: NEWS Layout
+1 On Tue, Feb 2, 2016 at 1:09 PM, Bill Farnerwrote: > +1 > > On Tuesday, February 2, 2016, Erb, Stephan > wrote: > > > Hi everyone, > > > > I'd like to propose that we give our NEWS file a little bit more > > structure. Currently, it is quite cluttered [1]. > > > > To keep it simple, I'd suggest that we adopt the style from the 0.11 > > Aurora blog post: > > > > * New/updated > > * Deprecations and removals > > > > [1] https://github.com/apache/aurora/blob/master/NEWS > > [2] https://aurora.apache.org/blog/aurora-0-11-0-released/ > > > > > > Thoughts? > > >
Re: NEWS Layout
+1 On Tue, Feb 2, 2016 at 11:25 AM, Jake Farrellwrote: > sounds good. thanks Stephan > > -Jake > > On Tue, Feb 2, 2016 at 2:05 PM, Erb, Stephan > wrote: > > > Hi everyone, > > > > I'd like to propose that we give our NEWS file a little bit more > > structure. Currently, it is quite cluttered [1]. > > > > To keep it simple, I'd suggest that we adopt the style from the 0.11 > > Aurora blog post: > > > > * New/updated > > * Deprecations and removals > > > > [1] https://github.com/apache/aurora/blob/master/NEWS > > [2] https://aurora.apache.org/blog/aurora-0-11-0-released/ > > > > > > Thoughts? > > >
Re: NEWS Layout
sounds good. thanks Stephan -Jake On Tue, Feb 2, 2016 at 2:05 PM, Erb, Stephanwrote: > Hi everyone, > > I'd like to propose that we give our NEWS file a little bit more > structure. Currently, it is quite cluttered [1]. > > To keep it simple, I'd suggest that we adopt the style from the 0.11 > Aurora blog post: > > * New/updated > * Deprecations and removals > > [1] https://github.com/apache/aurora/blob/master/NEWS > [2] https://aurora.apache.org/blog/aurora-0-11-0-released/ > > > Thoughts? >
Re: NEWS Layout
Thanks for the quick responses! Now on master: https://reviews.apache.org/r/43109/ From: Zameer ManjiSent: Tuesday, February 2, 2016 8:35 PM To: dev@aurora.apache.org Subject: Re: NEWS Layout +1 On Tue, Feb 2, 2016 at 11:32 AM, John Sirois wrote: > +1 > > On Tue, Feb 2, 2016 at 12:29 PM, Joshua Cohen wrote: > > > +1 > > > > On Tue, Feb 2, 2016 at 1:09 PM, Bill Farner wrote: > > > > > +1 > > > > > > On Tuesday, February 2, 2016, Erb, Stephan < > stephan@blue-yonder.com> > > > wrote: > > > > > > > Hi everyone, > > > > > > > > I'd like to propose that we give our NEWS file a little bit more > > > > structure. Currently, it is quite cluttered [1]. > > > > > > > > To keep it simple, I'd suggest that we adopt the style from the 0.11 > > > > Aurora blog post: > > > > > > > > * New/updated > > > > * Deprecations and removals > > > > > > > > [1] https://github.com/apache/aurora/blob/master/NEWS > > > > [2] https://aurora.apache.org/blog/aurora-0-11-0-released/ > > > > > > > > > > > > Thoughts? > > > > > > > > > > > > > -- > John Sirois > 303-512-3301 > > -- > Zameer Manji > > <303-512-3301>
Re: Rollback Testing
https://issues.apache.org/jira/browse/AURORA-1608 https://issues.apache.org/jira/browse/AURORA-1609 On Tue, Feb 2, 2016 at 7:42 PM, Joshua Cohenwrote: > Ok, I'll file a ticket to at least track the intent. We'll see if I have > time to work on it or if it just languishes (anyone else interested, feel > free to pick it up as well of course ;)). > > On Mon, Feb 1, 2016 at 4:28 PM, Bill Farner wrote: > >> Yes. Our nightly packaging builds take advantage of this. >> >> On Mon, Feb 1, 2016 at 2:27 PM, Joshua Cohen wrote: >> >> > I know the blocker for e2e tests in CI previously was the ability to run >> > vagrant. Is running docker from CI doable today? >> > >> > On Mon, Feb 1, 2016 at 2:40 PM, Bill Farner wrote: >> > >> > > Definitely a nice thing to have, the big uncertainty will be whether >> > anyone >> > > cares enough to see the effort through. >> > > >> > > e2e tests in jenkins can be done, but likely only if e2e tests start >> > using >> > > docker instead of vagrant. I am supportive of both, but cannot >> > personally >> > > invest the time at the moment. >> > > >> > > On Mon, Feb 1, 2016 at 12:14 PM, Joshua Cohen >> wrote: >> > > >> > > > Another topic that came up today's IRC meeting was possibly adding >> some >> > > > sort of automated rollback testing between builds. This is related >> to >> > > > https://issues.apache.org/jira/browse/AURORA-1603 which was caused >> by >> > an >> > > > inability to rollback to an earlier commit after discovering >> > > > https://issues.apache.org/jira/browse/AURORA-1605. >> > > > >> > > > I'm not sure what's feasible in the immediate future. To me the most >> > > > obvious solution would be to run the e2e tests as part of our CI >> job, >> > and >> > > > have the last step of the job be to checkout HEAD^ and >> rebuild/restart >> > > the >> > > > scheduler. This is predicated on actually running the e2e tests as >> part >> > > of >> > > > CI though, which historically has been difficult for us to achieve >> > (see: >> > > > https://issues.apache.org/jira/browse/AURORA-127). >> > > > >> > > > Curious to hear thoughts on this. It's clearly more beneficial to >> those >> > > > deploying from master rather than from official releases (though it >> > would >> > > > be great, at least, as part of the release verification process to >> > ensure >> > > > we can rollback to the previous release). Do folks think this would >> be >> > > > beneficial, or is it overkill? >> > > > >> > > > Cheers, >> > > > >> > > > Joshua >> > > > >> > > >> > >> > >
Re: Rollback Testing
Ok, I'll file a ticket to at least track the intent. We'll see if I have time to work on it or if it just languishes (anyone else interested, feel free to pick it up as well of course ;)). On Mon, Feb 1, 2016 at 4:28 PM, Bill Farnerwrote: > Yes. Our nightly packaging builds take advantage of this. > > On Mon, Feb 1, 2016 at 2:27 PM, Joshua Cohen wrote: > > > I know the blocker for e2e tests in CI previously was the ability to run > > vagrant. Is running docker from CI doable today? > > > > On Mon, Feb 1, 2016 at 2:40 PM, Bill Farner wrote: > > > > > Definitely a nice thing to have, the big uncertainty will be whether > > anyone > > > cares enough to see the effort through. > > > > > > e2e tests in jenkins can be done, but likely only if e2e tests start > > using > > > docker instead of vagrant. I am supportive of both, but cannot > > personally > > > invest the time at the moment. > > > > > > On Mon, Feb 1, 2016 at 12:14 PM, Joshua Cohen > wrote: > > > > > > > Another topic that came up today's IRC meeting was possibly adding > some > > > > sort of automated rollback testing between builds. This is related to > > > > https://issues.apache.org/jira/browse/AURORA-1603 which was caused > by > > an > > > > inability to rollback to an earlier commit after discovering > > > > https://issues.apache.org/jira/browse/AURORA-1605. > > > > > > > > I'm not sure what's feasible in the immediate future. To me the most > > > > obvious solution would be to run the e2e tests as part of our CI job, > > and > > > > have the last step of the job be to checkout HEAD^ and > rebuild/restart > > > the > > > > scheduler. This is predicated on actually running the e2e tests as > part > > > of > > > > CI though, which historically has been difficult for us to achieve > > (see: > > > > https://issues.apache.org/jira/browse/AURORA-127). > > > > > > > > Curious to hear thoughts on this. It's clearly more beneficial to > those > > > > deploying from master rather than from official releases (though it > > would > > > > be great, at least, as part of the release verification process to > > ensure > > > > we can rollback to the previous release). Do folks think this would > be > > > > beneficial, or is it overkill? > > > > > > > > Cheers, > > > > > > > > Joshua > > > > > > > > > >
Re: NEWS Layout
+1 On Tuesday, February 2, 2016, Erb, Stephanwrote: > Hi everyone, > > I'd like to propose that we give our NEWS file a little bit more > structure. Currently, it is quite cluttered [1]. > > To keep it simple, I'd suggest that we adopt the style from the 0.11 > Aurora blog post: > > * New/updated > * Deprecations and removals > > [1] https://github.com/apache/aurora/blob/master/NEWS > [2] https://aurora.apache.org/blog/aurora-0-11-0-released/ > > > Thoughts? >
Re: NEWS Layout
+1 On Tue, Feb 2, 2016 at 12:29 PM, Joshua Cohenwrote: > +1 > > On Tue, Feb 2, 2016 at 1:09 PM, Bill Farner wrote: > > > +1 > > > > On Tuesday, February 2, 2016, Erb, Stephan > > wrote: > > > > > Hi everyone, > > > > > > I'd like to propose that we give our NEWS file a little bit more > > > structure. Currently, it is quite cluttered [1]. > > > > > > To keep it simple, I'd suggest that we adopt the style from the 0.11 > > > Aurora blog post: > > > > > > * New/updated > > > * Deprecations and removals > > > > > > [1] https://github.com/apache/aurora/blob/master/NEWS > > > [2] https://aurora.apache.org/blog/aurora-0-11-0-released/ > > > > > > > > > Thoughts? > > > > > > -- John Sirois 303-512-3301
0.12.0 RC status
Although the last blocker raised for the 0.12.0 RC series has been resolved [1], it looks like resolution of several issues related to rolling back to 0.11.0 are required to cut the next RC: 1. "Scheduler fails to start after rollback": https://issues.apache.org/jira/browse/AURORA-1603 2. "Add a flag to disable the HTTP redirect to the leader": https://issues.apache.org/jira/browse/AURORA-1601 3. "Update recovery docs to reflect changes": https://issues.apache.org/jira/browse/AURORA-1605 These issues fall into 2 classes: Item 1 above needs to fix the immediate problem of rolling back to 0.11.0; although there may be more changes to process, tooling and code to support the problem better going forward. Items 2 & 3 address tooling & procedure that support rollback. It looks like Maxim has claimed item 1/AURORA-1603 and Joshua is working item 2/AURORA-1601. I assume one of Maxim, Joshua or Zameer will tackle item 3/AURORA-1605 to update rollback docs with what they learned rolling back. If I have any of this wrong, please speak up; otherwise I'll be cutting the next 0.12.0 RC3 when the above 3 issues are resolved. [1] "Identity.role is still used in the UI leading to duplicate instances on job page": https://issues.apache.org/jira/browse/AURORA-1604
Re: [PROPOSAL] Change java thrift code gen
To wrap things up - I'm considering this proposal as having failed; so I've marked the 3 associated RBs as discarded. I'll be working on https://issues.apache.org/jira/browse/THRIFT-3583 to introduce an immutable builder-style of java codegen that carries over thrift annotations to Apache Thrift. Once that is complete and released, I'll circle back and re-do an RB like 3/3 ( https://reviews.apache.org/r/42756/) that flips over the codebase to the new apache thrift gen. On Thu, Jan 28, 2016 at 10:08 AM, John Siroiswrote: > > > On Thu, Jan 28, 2016 at 8:26 AM, John Sirois wrote: > >> On Thu, Jan 28, 2016 at 6:45 AM, Jake Farrell >> wrote: >> >> > +1 to making this apart of Thrift, i'm happy to help shepard this on the >> > Thrift side and get it in as soon as its ready >> > >> >> I've filed https://issues.apache.org/jira/browse/THRIFT-3583 to use as >> the >> basis for discussion of this feature over in the Apache Thrift project. >> I looked at the problem a bit and noted some challenges. >> > > And started a d...@thrift.apache.org thread here if anyone wants to follow > along or participate: http://markmail.org/message/mlxyyauyvlvuxjsf > > >> >> >> > >> > -Jake >> > >> > On Wed, Jan 27, 2016 at 8:08 PM, Maxim Khutornenko >> > wrote: >> > >> >> I am +1 to making immutable thrift objects solely based on perf >> numbers. >> >> >> >> My biggest concern though is maintenance of a pretty intricate >> codebase, >> >> especially when it comes to upgrading any of the frameworks involved. >> >> Bill's suggestion to explore paths to make this a part of Apache Thrift >> >> sounds great to me. >> >> >> >> On Wed, Jan 27, 2016 at 5:00 PM, Bill Farner >> wrote: >> >> >> >> > Tony - this would not be a technical fork so much as a spiritual >> fork. >> >> > While we would have our own bugs to work out, the only upstream >> exposure >> >> > would be IDL or wire format changes. >> >> > >> >> > On Wed, Jan 27, 2016 at 4:58 PM, Tony Dong > > >> >> > wrote: >> >> > >> >> > > Awesome performance numbers! I don't particularly know the >> logistics >> >> of >> >> > > upstreaming a change like this, but optimistically I would suggest >> >> > > upstreaming it to Apache Thrift if possible. >> >> > > >> >> > > As someone that maintains a fork of a thrift compiler(fork of >> >> scrooge), I >> >> > > have to say that it's not that fun. There's a lot of custom code >> that >> >> > needs >> >> > > to be maintained and a bunch of work to rebase the code >> periodically. >> >> > > >> >> > > >> >> > > >> >> > > On Wed, Jan 27, 2016 at 4:51 PM, Bill Farner >> >> wrote: >> >> > > >> >> > > > Firstly - thanks for the clean organization and delineation of >> >> steps in >> >> > > > this change. Top notch work! >> >> > > > >> >> > > > Some of the performance improvements are very nice; and in a >> >> > particularly >> >> > > > hot code path. I will wager a guess that the majority of the >> >> savings >> >> > is >> >> > > in >> >> > > > avoiding what amounts to copy constructors between mutable and >> >> > immutable >> >> > > > types. I further wager there are alternative approaches we could >> >> weigh >> >> > > to >> >> > > > achieve those performance improvements. As an example - you note >> >> above >> >> > > > that we could provide a patch to Apache Thrift. Depending how >> much >> >> > > > performance inspires our decision here, it will be prudent to >> >> evaluate >> >> > > > alternatives. >> >> > > > >> >> > > > I think there are (at least) two major issues worth discussing - >> >> code >> >> > > > volume (which you note) and an increase in logical complexity. >> This >> >> > will >> >> > > > leave us with a bifurcation in code generation tooling >> (custom+swift >> >> > for >> >> > > > Java, Apache Thrift for python and js). It's difficult to >> quantify >> >> the >> >> > > > downside of that, but it seems like an unfortunate state with >> >> potential >> >> > > for >> >> > > > compatibility risks. >> >> > > > >> >> > > > >> >> > > > On Wed, Jan 27, 2016 at 2:45 PM, Zameer Manji > > >> >> > wrote: >> >> > > > >> >> > > > > Some high level comments without looking at the code. >> >> > > > > >> >> > > > > I'm in favor from abandoning the thrift generated java code in >> >> favor >> >> > of >> >> > > > > immutable objects. I think it is easier to reason about and >> will >> >> > ensure >> >> > > > we >> >> > > > > have less errors in our code. If I understand correctly, the >> >> ProtoBuf >> >> > > > > format does this by default, so there some precedent for this >> >> style >> >> > of >> >> > > > code >> >> > > > > generation already. >> >> > > > > >> >> > > > > I think using Facebook's swift is the best approach here. I >> would >> >> be >> >> > > > > hesitant to accept any custom code generation that involved us >> >> > parsing >> >> > > > > thrift IDL
Re: 0.12.0 RC status
+1 to having 1603 and 1601 as blockers. I am planning to work on 1603 today. As for 1605, I don't believe it's a blocker given that all findings are already documented in the ticket. On Tue, Feb 2, 2016 at 7:03 AM, Joshua Cohenwrote: > I'd only consider item 1 to be a blocker to 0.12.0, but 2 and 3 should be > relatively quick so in general this sounds like a reasonable plan of action > to me. > > On Tue, Feb 2, 2016 at 8:52 AM, John Sirois wrote: > > > Although the last blocker raised for the 0.12.0 RC series has been > resolved > > [1], it looks like resolution of several issues related to rolling back > to > > 0.11.0 are required to cut the next RC: > > 1. "Scheduler fails to start after rollback": > > https://issues.apache.org/jira/browse/AURORA-1603 > > 2. "Add a flag to disable the HTTP redirect to the leader": > > https://issues.apache.org/jira/browse/AURORA-1601 > > 3. "Update recovery docs to reflect changes": > > https://issues.apache.org/jira/browse/AURORA-1605 > > > > These issues fall into 2 classes: > > Item 1 above needs to fix the immediate problem of rolling back to > 0.11.0; > > although there may be more changes to process, tooling and code to > support > > the problem better going forward. > > Items 2 & 3 address tooling & procedure that support rollback. > > > > It looks like Maxim has claimed item 1/AURORA-1603 and Joshua is working > > item 2/AURORA-1601. I assume one of Maxim, Joshua or Zameer will tackle > > item 3/AURORA-1605 to update rollback docs with what they learned rolling > > back. > > > > If I have any of this wrong, please speak up; otherwise I'll be cutting > the > > next 0.12.0 RC3 when the above 3 issues are resolved. > > > > [1] "Identity.role is still used in the UI leading to duplicate instances > > on job page": https://issues.apache.org/jira/browse/AURORA-1604 > > >