But the first line should also be below 50 characters [1]. That might
become tricky with a single component and it becomes more difficult, if
a change touches more than one component (the ellipsis in [2] show the
reason).
To include components into the commit message it might be better to do
that in the body instead of the subject (and we might want to use the
same name that we use in JIRA).
Also, it does seem redundant to have the components that are available
in the referenced JIRA issue again in the message, but then it’s not
exactly trivial to join the log messages with the components in JIRA
(unless both are available in AsterixDB ;) )
Alternatively, we could think about really short acronyms for the
components to make them fit.
Thoughts?
Till
[1] https://cwiki.apache.org/confluence/display/ASTERIXDB/Formatting
[2] https://github.com/apache/spark/commits/master
On 15 Jun 2017, at 14:55, Mike Carey wrote:
+1
On 6/15/17 1:19 PM, Yingyi Bu wrote:
Each commit message should
1) reference 1 or more JIRA issues (that hopefully provide a
rationale
for
the change).
2) contain a description of changes to the user model (language
syntax,
configuration parameters, ..)
3) contain a description of storage format changes (that would
require
reloading or upgrading)
4) contain a description of interface changes (for source code
consumers)
I wonder if we could put component name(s) into the headline of
commit
message.
So the headline will be sth. like:
[ASTERIXDB-2000][TXN] Fix a deadlock in LogManager
[ASTERIXDB-2001][STORAGE] Clean up file handling in BufferCache
....
It makes change localization easier.
Examples:
Spark: https://github.com/apache/spark/commits/master
Flink: https://github.com/apache/flink/commits/master
Here is a list of component names:
- API
- AQL
- CLUSTER (cluster management)
- COMPILER
- CONFIGURATION (configuration parameters/mgmt)
- DOC
- FAILURE (failure handling)
- EXTERNAL
- INDEX
- INGESTION
- FUNC (function)
- LICENSE
- MAVEN
- METADATA
- PERF (performance monitoring etc.)
- REPLICATION
- RPC (cc/nc message)
- RUNTIME
- STATS (statistics etc.)
- SITE
- STORAGE
- SQL++
- TEST
- TXN (transaction)
- TYPE (data model)
- UDF (user defined function)
- UI
Best,
Yingyi
On Thu, Jun 15, 2017 at 1:09 AM, Yingyi Bu <[email protected]>
wrote:
+1!
Best,
Yingyi
On Wed, Jun 14, 2017 at 10:43 PM, Mike Carey <[email protected]>
wrote:
+1 !!!
I think this is a GREAT proposal, and we can also then hopefully do
the
equivalent of grep'ing the commits to identify things that we might
want to
incorporate in a high-level set of release notes. I also really
like the
"no" requirement. Also, storage changes should really NOT be taken
lightly
- they seriously inconvenience our customers and will hopefully
cause the
tests to fail (the ones that check for backward-compat) - I would
like to
set storage changes actually be voted on, ideally, if we could make
that
part of our operating procedure somehow?
Cheers,
Mike
On 6/14/17 9:15 AM, Till Westmann wrote:
Hi,
some of us had a discussion with an SDSC team last week that is
running
an
AsterixDB instance. Their customer perspective on AsterixDB
highlighted a
few areas of improvement to ease the consumption of AsterixDB.
One thing that I’d like to follow up on are release notes. So
far we
didn’t
provide them and so changes to the system came as a surprise to
everybody
who is not monitoring commits closely.
As I think that it’s not easy to provide good release notes
I’d like to
propose some additions to our commit messages to ease the creation
of
release messages:
Each commit message should
1) reference 1 or more JIRA issues (that hopefully provide a
rationale
for
the change).
2) contain a description of changes to the user model (language
syntax,
configuration parameters, ..)
3) contain a description of storage format changes (that would
require
reloading or upgrading)
4) contain a description of interface changes (for source code
consumers)
and all reviewers should check that these are mentioned in the
commit
message. To increase the probability to we won’t forget to
mention the
changes in 2-4, I think that we should should explicitly mention
the
absence
of such changes, i.e.:
user model changes: no
storage format changes: no
interface changes: no
Thoughts/concerns about this?
Is this manageable?
Are there other kinds of changes that have a high impact on
consumers
that
we should call out?
Cheers,
Till´
