About IMPALA-9690 (AVX support): My preferred solution would be to deprecate support for x64 without AVX2 in 4.0, but not start removing the related logic yet. - We could even add a DCHECK + flag to crash by default if no AVX2 is detected, and a message that points them to Impala mailing lists/related Jira. - After 4.0 is released people would be free to implement things that assume AVX2 support on x64, for example compiling Impala/the whole toolchain with -mavx2. I think that currently people are held back from doing this - why put effort in doing+benchmarking a change if you can't push it to master. - If a user appears with a valid use case that needs x64 without AVX2 then we can revise our plans, but we shouldn't ignore low hanging fruits for the sake of imaginary users :)
-Csaba On Sun, Apr 11, 2021 at 5:52 PM Shant Hovsepian <sh...@superdupershant.com> wrote: > +1 > > Regarding IMPALA-9690 if someone has the time to clean up the old > conditional cpu logic it would be great, but if the ARM folks still > need some of that logic it's not critical enough to block, just a very > nice to have? > > -Shant > > On Sun, Apr 4, 2021 at 1:09 AM Tim Armstrong <tim.g.armstr...@gmail.com> > wrote: > > > > I agree that it would be good to get this done even if it's not perfect, > > there's so much good stuff. > > > > On Fri, 2 Apr 2021 at 17:40, Joe McDonnell <joemcdonn...@cloudera.com> > > wrote: > > > > > I agree that we should wrap up Impala 4. A large amount of good work > has > > > gone in and it belongs in a release. > > > > > > We need to decide which breaking changes are truly blockers for an > Impala 4 > > > release. My feeling is that we can't hold the release for compatibility > > > breaking changes unless someone signs up to do those changes. Breaking > > > compatibility is useful sometimes, but it can't be an indefinite hold > on > > > releases. I think a release discussion thread is a good way to > stimulate > > > this discussion. > > > > > > Apart from the two JIRAs you listed, one other change mentioned in the > > > original email is switching to use the new on-demand metadata by > default. > > > > > > Thanks, > > > Joe > > > > > > On Fri, Apr 2, 2021 at 5:22 PM Quanlong Huang <huangquanl...@gmail.com > > > > > wrote: > > > > > > > Sure. If there are no objections, I'll raise a discussion thread for > the > > > > 4.0 release. There are still some unresolved breaking changes, e.g. > > > > > > > > - IMPALA-2210: Make Parquet the default file format > > > > - IMPALA-9690: Bump minimum x86-64 CPU requirements > > > > > > > > Thanks > > > > Quanlong > > > > > > > > On Tue, Mar 30, 2021 at 11:13 PM Jim Apple <jbap...@apache.org> > wrote: > > > > > > > > > Thanks for bringing this up again, Quanlong! I would love to see a > 4.0 > > > > > release soon so that 4.1 releases and what not can be prepared. > Already > > > > the > > > > > changlist from the 3.x line is going to be enormous and could be > hard > > > for > > > > > users to digest. > > > > > > > > > > On Mon, Mar 29, 2021 at 7:22 AM Quanlong Huang < > > > huangquanl...@gmail.com> > > > > > wrote: > > > > > > > > > > > Reviving this thread. I think it's time to create the 4.0.0 > branch > > > and > > > > > > prepare for the release now. Any breaking change is landing and > we > > > > should > > > > > > wait? > > > > > > > > > > > > Thanks, > > > > > > Quanlong > > > > > > > > > > > > On Thu, Jun 18, 2020 at 5:47 AM Tim Armstrong < > > > tarmstr...@cloudera.com > > > > > > > > > > > wrote: > > > > > > > > > > > > > Another thing that we're looking into is changing the encoded > > > runtime > > > > > > > profile representation to be more efficient - see > > > > > > > https://issues.apache.org/jira/browse/IMPALA-9378. So that > might > > > be > > > > a > > > > > > > default we will try to change in Impala 4.0. > > > > > > > > > > > > > > On Fri, Apr 24, 2020 at 12:37 PM Tim Armstrong < > > > > > tarmstr...@cloudera.com> > > > > > > > wrote: > > > > > > > > > > > > > > > An addendum to this - we're also considering whether to > increase > > > > the > > > > > > > > minimum CPU version so that we can simplify development and > focus > > > > > more > > > > > > on > > > > > > > > optimising for the most common CPUs - see > > > > > > > > https://issues.apache.org/jira/browse/IMPALA-9690 > > > > > > > > > > > > > > > > Running on commodity hardware is important and we want > people to > > > be > > > > > > able > > > > > > > > to develop on whatever machine they have access to. At most > we're > > > > > > likely > > > > > > > > going to require AVX2 support, which has been supported by > the > > > vast > > > > > > > > majority of CPUs for the better part of a decade. > > > > > > > > > > > > > > > > On Tue, Apr 7, 2020 at 11:50 AM Joe McDonnell < > > > > > > joemcdonn...@cloudera.com > > > > > > > > > > > > > > > > wrote: > > > > > > > > > > > > > > > >> Impala is starting development on Impala 4.0. There are many > > > > > exciting > > > > > > > >> projects underway including full support for Hive ACID > tables, > > > > > > improved > > > > > > > >> multithreading support, and fault tolerance improvements, > but > > > > Impala > > > > > > 4.0 > > > > > > > >> also serves as a time to alter behavior and platform > support. > > > Here > > > > > is > > > > > > > the > > > > > > > >> current list of notable breaking changes that are planned in > > > > Impala > > > > > > 4.0: > > > > > > > >> > > > > > > > >> Remove support for older operating systems: > > > > > > > >> > > > > > > > >> Maintaining support for older operating systems requires a > > > > > > considerable > > > > > > > >> amount of effort, especially with changing security > > > requirements. > > > > > Each > > > > > > > >> extra operating system consumes valuable effort and can > conflict > > > > > with > > > > > > > >> support for newer operating systems, so we have decided to > > > remove > > > > > > > support > > > > > > > >> for some older operating systems. Specifically, we plan to > drop > > > > > > support > > > > > > > for: > > > > > > > >> > > > > > > > >> - > > > > > > > >> > > > > > > > >> Centos 6 > > > > > > > >> - > > > > > > > >> > > > > > > > >> Ubuntu 14 > > > > > > > >> - > > > > > > > >> > > > > > > > >> Debian 8 > > > > > > > >> > > > > > > > >> Each of these is approaching or past its end of life. Since > > > > Centos6 > > > > > > was > > > > > > > >> the only operating system still needing Python 2.6 and > Python > > > 2.6 > > > > is > > > > > > > long > > > > > > > >> past its end of life, Python 2.6 will no longer be > supported. In > > > > > > > addition, > > > > > > > >> Centos 7 support will be focused on newer versions such as > 7.5 > > > and > > > > > > > above. > > > > > > > >> > > > > > > > >> Remove support for Sentry: > > > > > > > >> > > > > > > > >> Over the past year and a half, Impala's Ranger authorization > > > > > > > >> functionality has achieved parity and surpassed the existing > > > > Sentry > > > > > > > >> authorization functionality. Impala's Sentry support > requires > > > > > ongoing > > > > > > > >> effort to test and maintain, including maintenance on Sentry > > > > itself > > > > > to > > > > > > > >> address security updates. Given the diminished activity in > the > > > > > Sentry > > > > > > > >> community, there is no timeline for addressing these > security > > > > > updates. > > > > > > > >> Unless there is a maintainer for Sentry support, Impala > plans to > > > > > focus > > > > > > > its > > > > > > > >> efforts on Ranger for its authorization system and drop > Sentry. > > > > > > > >> > > > > > > > >> Remove support for Impala-lzo: > > > > > > > >> > > > > > > > >> Impala-lzo provides code to allow Impala to read the LZO > > > > compressed > > > > > > > >> tables. LZO is GPL licensed, which is why this support is > not > > > > > included > > > > > > > >> directly. The Impala-lzo code interacts with internal Impala > > > code > > > > > at a > > > > > > > >> level that is error prone and intricate. Given the low > adoption > > > of > > > > > LZO > > > > > > > and > > > > > > > >> the other compression options available, Impala plans to > remove > > > > > > > Impala-lzo > > > > > > > >> support along with the low level interface it used. > > > > > > > >> > > > > > > > >> Deprecations: > > > > > > > >> > > > > > > > >> In addition, we also plan to deprecate several existing > features > > > > so > > > > > > that > > > > > > > >> they can be removed in a future release. Here is a summary > of > > > the > > > > > most > > > > > > > >> notable ones: > > > > > > > >> > > > > > > > >> - > > > > > > > >> > > > > > > > >> Impala will default to using on demand metadata > management > > > > (i.e. > > > > > > the > > > > > > > >> local catalog implementation) as described in this > > > > documentation: > > > > > > > >> > > > > > > > http://impala.apache.org/docs/build/html/topics/impala_metadata.html > > > > > > > >> The old metadata system is now deprecated and may be > removed > > > > in a > > > > > > > >> future release. > > > > > > > >> - > > > > > > > >> > > > > > > > >> Impala is deprecating the Beeswax client protocol (i.e. > > > clients > > > > > > that > > > > > > > >> connect via beeswax_port) in favor of the HiveServer2 > client > > > > > > > protocol. > > > > > > > >> - > > > > > > > >> > > > > > > > >> Impala is deprecating the old decimal_v2=false behavior > from > > > > > Impala > > > > > > > >> 2.x. The decimal_v2 query option has defaulted to true > since > > > > > Impala > > > > > > > 3. > > > > > > > >> > > > > > > > >> > > > > > > > >> Impala is always open to new developers, and we welcome > feedback > > > > on > > > > > > > these > > > > > > > >> plans. Further discussion of other changes and deprecations > is > > > > > ongoing > > > > > > > at > > > > > > > >> dev@impala.apache.org. > > > > > > > >> > > > > > > > >> > > > > > > > >> Thanks, > > > > > > > >> > > > > > > > >> Joe McDonnell > > > > > > > >> > > > > > > > >> > > > > > > > > > > > > > > > > > > > > > > > > > >
