Hi everyone,
I have create an issue [1] to track the progress of this topic. I have
written a little design document [2] how we could implement the
indicators and which parts have to be touched. I would suggest to
implement a prototype, also to see what is possible and can be
integrated both in Flink and Calcite. Feedback is welcome.
Regards,
Timo
[1] https://issues.apache.org/jira/browse/FLINK-5884
[2]
https://docs.google.com/document/d/1JRXm09x_wKst6z6UXdCGF9tgF1ueOAsFiQwahR72vbc/edit?usp=sharing
Am 21/02/17 um 15:06 schrieb Fabian Hueske:
Hi Xingcan,
thanks for your thoughts.
In principle you are right that the monotone attribute property would be
sufficient, however there are more aspects to consider than that.
Flink is a parallel stream processor engine which means that data is
processed in separate processes and shuffle across them.
Maintaining a strict order when merging parallel streams would be
prohibitively expensive.
Flink's watermark mechanism helps operators to deal with out-of-order data
(due to out-of-order input or shuffles).
I don't think we can separate the discussion about time attributes from
watermarks if we want to use Flink as a processing engine and not
reimplement large parts from scratch.
When transforming a time attribute, we have to either align it with
existing watermarks or generate new watermarks.
If we want to allow all kinds of monotone transformations, we have to adapt
the watermarks which is not trivial.
Instead, I think we should initially only allow very few monotone
transformations which are aligned with the existing watermarks. We might
later relax this condition if we see that users request this feature.
You are right, that we need to track which attribute can be used as a time
attribute (i.e., is increasing and guarded by watermarks).
For that we need to expose the time attribute when a Table is created
(either when a DataStream is converted like: stream.toTable(tEnv, 'a, 'b,
't.rowtime) or in a StreamTableSource) and track how it is used in queries.
I am not sure if the monotone property would be the right choice here,
since data is only quasi-monotone and a monotone annotation might trigger
some invalid optimizations which change the semantics of a query.
Right now, Calcite does not offer a quasi-monotone property (at least I
haven't found it).
Best, Fabian
2017-02-21 4:41 GMT+01:00 Xingcan Cui <xingc...@gmail.com>:
Hi all,
As I said in another thread, the main difference between stream and table
is that a stream is an ordered list while a table is an unordered set.
Without considering the out-of-order problem in practice, whether
event-time or processing-time can be just taken as a monotonically
increasing field and that's why the given query[1] would work. In other
words, we must guarantee the "SELECT MAX(t22.rowtime) ..." subquery returns
a single value that can be retrieved from the cached dynamic table since
it's dangerous to join two un-windowed streams.
Under this circumstance, I just consider adding a "monotonic hint"(INC or
DEC) to the field of a (generalized) table (maybe using an annotation on
the registerDataXX method) that can be used to indicate whether a field is
monotonically increasing or decreasing. Then by taking rowtime as common
(monotonically increasing) field, there are several benefits:
1) This can uniform the table and stream by importing total ordering
relation to an unordered set.
2) These fields can be modified arbitrarily as long as they keep the
declared monotonic feature and the watermark problem does not exist any
more.
3) The monotonic hint will be useful in the query optimization process.
What do you think?
Best,
Xingcan
[1]
SELECT t1.amount, t2.rate
FROM
table1 AS t1,
table2 AS t2
WHERE
t1.currency = t2.currency AND
t2.rowtime = (
SELECT MAX(t22.rowtime)
FROM table2 AS t22
AND t22.rowtime <= t1.rowtime)
On Tue, Feb 21, 2017 at 2:52 AM, Fabian Hueske <fhue...@gmail.com> wrote:
Hi everybody,
When Timo wrote to the Calcite mailing list, Julian Hyde replied and gave
good advice and explained why a system attribute for event-time would be
a
problem [1].
I thought about this and agree with Julian.
Here is a document to describe the problem, constraints in Flink and a
proposal how to handle processing time and event time in Table API and
SQL:
->
https://docs.google.com/document/d/1MDGViWA_
TCqpaVoWub7u_GY4PMFSbT8TuaNl-
EpbTHQ
Please have a look, comment and ask questions.
Thank you,
Fabian
[1]
https://lists.apache.org/thread.html/6397caf0ca37f97f2cd27d96f7a12c
6fa845d6fd0870214fdce18d96@%3Cdev.calcite.apache.org%3E
2017-02-16 1:18 GMT+01:00 Fabian Hueske <fhue...@gmail.com>:
Thanks everybody for the comments.
Actually, I think we do not have much choice when deciding whether to
use
attributes or functions.
Consider the following join query:
SELECT t1.amount, t2.rate
FROM
table1 AS t1,
table2 AS t2
WHERE
t1.currency = t2.currency AND
t2.rowtime = (
SELECT MAX(t22.rowtime)
FROM table2 AS t22
AND t22.rowtime <= t1.rowtime)
The query joins two streaming tables. Table 1 is a streaming table with
amounts in a certain currency. Table 2 is a (slowly changing) streaming
table of currency exchange rates.
We want to join the amounts stream with the exchange rate of the
corresponding currency that is valid (i.e., last received value ->
MAX(rowtime)) at the rowtime of the amounts row.
In order to specify the query, we need to refer to the rowtime of the
different tables. Hence, we need a way to relate the rowtime expression
(or
marker) to a table.
This is not possible with a parameterless scalar function.
I'd like to comment on the concerns regarding the performance:
In fact, the columns could be completely virtual and only exist during
query parsing and validation.
During execution, we can directly access the rowtime metadata of a
Flink
streaming record (which is present anyway) or look up the current
processing time from the machine clock. So the processing overhead
would
actually be the same as with a marker function.
Regarding the question on what should be allowed with a system
attribute:
IMO, it could be used as any other attribute. We need it at least in
GROUP
BY, ORDER BY, and WHERE to define windows and joins. We could also
allow
to
access it in SELECT if we want users to give access to rowtime and
processing time. So @Haohui, your query could be supported.
However, what would not be allowed is to modify the value of the rows,
i.e., by naming another column rowtime, i.e., "SELECT sometimestamp AS
rowtime" would not be allowed, because Flink does not support to modify
the
event time of a row (for good reasons) and processing time should not
be
modifiable anyway.
@Timo:
I think the approach to only use the system columns during parsing and
validation and converting them to expressions afterwards makes a lot of
sense.
The question is how this approach could be nicely integrated with
Calcite.
Best, Fabian
2017-02-15 16:50 GMT+01:00 Radu Tudoran <radu.tudo...@huawei.com>:
Hi,
My initial thought would be that it makes more sense to thave
procTime()
and rowTime() only as functions which in fact are to be used as
markers.
Having the value (even from special system attributes does not make
sense
in some scenario such as the ones for creating windows, e.g.,
If you have SELECT Count(*) OVER (ORDER BY procTime()...)
If you get the value of procTime you cannot do anything as you need
the
marker to know how to construct the window logic.
However, your final idea of having " implement some rule/logic that
translates the attributes to special RexNodes internally " I believe
is
good and gives a solution to both problems. One the one hand for those
scenarios where you need the value you can access the value, while for
others you can see the special type of the RexNode and use it as a
marker.
Regarding keeping this data in a table...i am not sure as you would
say
we need to augment the data with two fields whether needed or
not...this
is nto necessary very efficient
Dr. Radu Tudoran
Senior Research Engineer - Big Data Expert
IT R&D Division
HUAWEI TECHNOLOGIES Duesseldorf GmbH
European Research Center
Riesstrasse 25, 80992 München
E-mail: radu.tudo...@huawei.com
Mobile: +49 15209084330
Telephone: +49 891588344173
HUAWEI TECHNOLOGIES Duesseldorf GmbH
Hansaallee 205, 40549 Düsseldorf, Germany, www.huawei.com
Registered Office: Düsseldorf, Register Court Düsseldorf, HRB 56063,
Managing Director: Bo PENG, Wanzhou MENG, Lifang CHEN
Sitz der Gesellschaft: Düsseldorf, Amtsgericht Düsseldorf, HRB 56063,
Geschäftsführer: Bo PENG, Wanzhou MENG, Lifang CHEN
This e-mail and its attachments contain confidential information from
HUAWEI, which is intended only for the person or entity whose address
is
listed above. Any use of the information contained herein in any way
(including, but not limited to, total or partial disclosure,
reproduction,
or dissemination) by persons other than the intended recipient(s) is
prohibited. If you receive this e-mail in error, please notify the
sender
by phone or email immediately and delete it!
-----Original Message-----
From: Timo Walther [mailto:twal...@apache.org]
Sent: Wednesday, February 15, 2017 9:33 AM
To: dev@flink.apache.org
Subject: Re: [DISCUSS] Table API / SQL indicators for event and
processing time
Hi all,
at first I also thought that built-in functions (rowtime() and
proctime()) are the easiest solution. However, I think to be
future-proof
we should make them system attributes; esp. to relate them to a
corresponding table in case of multiple tables. Logically they are
attributes of each row, which is already done in Table API.
I will ask on the Calcite ML if there is a good way for integrating
system attributes. Right now, I would propose the following
implementation:
- we introduce a custom row type (extending RelDataType)
- in a streaming environment every row has two attributes by default
(rowtime and proctime)
- we do not allow creating a row type with those attributes (this
should
already prevent `SELECT field AS rowtime FROM ...`)
- we need to ensure that these attributes are not part of expansion
like
`SELECT * FROM ...`
- implement some rule/logic that translates the attributes to special
RexNodes internally, such that the opimizer does not modify these
attributes
What do you think?
Regards,
Timo
Am 15/02/17 um 03:36 schrieb Xingcan Cui:
Hi all,
thanks for this thread.
@Fabian If I didn't miss the point, the main difference between the
two approaches is whether or not taking these time attributes as
common table fields that are directly available to users. Whatever,
these time attributes should be attached to records (right?), and
the
discussion lies in whether give them public qualifiers like other
common fields or private qualifiers and related get/set methods.
The former (system attributes) approach will be more compatible with
existing SQL read-only operations (e.g., select, join), but we need
to
add restrictions on SQL modification operation (like what?). I think
there are no needs to forbid users modifying these attributes via
table APIs (like map function). Just inform them about these special
attribute names like system built in aggregator names in iteration.
As for the built in function approach, I don't know if, for now,
there
are functions applied on a single row (maybe the value access
functions like COMPOSITE.get(STRING)?). It seems that most of the
built in functions work for a single field or on columns and thus it
will be mountains of work if we want to add a new kind of function
to
SQL. Maybe all existing operations should be modified to support it.
All in all, if there are existing supports for single row function,
I
prefer the built in function approach. Otherwise the system
attributes
approach should be better. After all there are not so much
modification operations in SQL and maybe we can use alias to support
time attributes setting (just hypothesis, not sure if it's
feasible).
@Haohui I think the given query is valid if we add a aggregate
function to (PROCTIME()
- ROWTIME()) / 1000 and it should be executed efficiently.
Best,
Xingcan
On Wed, Feb 15, 2017 at 6:17 AM, Haohui Mai <ricet...@gmail.com>
wrote:
Hi,
Thanks for starting the discussion. I can see there are multiple
trade-offs in these two approaches. One question I have is that to
which extent Flink wants to open its APIs to allow users to access
both processing and event time.
Before we talk about joins, my understanding for the two approaches
that you mentioned are essentially (1) treating the value of event
/
processing time as first-class fields for each row, (2) limiting
the
scope of time indicators to only specifying windows. Take the
following query as an
example:
SELECT (PROCTIME() - ROWTIME()) / 1000 AS latency FROM table GROUP
BY
FLOOR(PROCTIME() TO MINUTES)
There are several questions we can ask:
(1) Is it a valid query?
(2) How efficient the query will be?
For this query I can see arguments from both sides. I think at the
end of the day it really comes down to what Flink wants to support.
After working on FLINK-5624 I'm more inclined to support the second
approach (i.e., built-in functions). The main reason why is that
the
APIs of Flink are designed to separate times from the real
payloads.
It probably makes sense for the Table / SQL APIs to have the same
designs.
For joins I don't have a clear answer on top of my head. Flink
requires two streams to be put in the same window before doing the
joins. This is essentially a subset of what SQL can express. I
don't
know what would be the best approach here.
Regards,
Haohui
On Tue, Feb 14, 2017 at 12:26 AM Fabian Hueske <fhue...@gmail.com>
wrote:
Hi,
It would as in the query I gave as an example before:
SELECT
a,
SUM(b) OVER (PARTITION BY c ORDER BY proctime ROWS BETWEEN 2
PRECEDING AND CURRENT ROW) AS sumB, FROM myStream
Here "proctime" would be a system attribute of the table
"myStream".
The table would also have another system attribute called
"rowtime"
which would be used to indicate event time semantics.
These attributes would always be present in tables which are
derived
from streams.
Because we still require that streams have timestamps and
watermarks
assigned (either by the StreamTableSource or the somewhere
downstream the DataStream program) when they are converted into a
table, there is no
need
to register anything.
Does that answer your questions?
Best, Fabian
2017-02-14 2:04 GMT+01:00 Radu Tudoran <radu.tudo...@huawei.com>:
Hi Fabian,
Thanks for starting the discussion. Before I give my thoughts on
this
can
you please give some examples of how would you see option of
using
"system
attributes"?
Do you use this when you register the stream as a table, do you
use
if when you call an SQL query, do you use it when you translate
back a
table
to a stream / write it to a dynamic table?
Dr. Radu Tudoran
Senior Research Engineer - Big Data Expert IT R&D Division
HUAWEI TECHNOLOGIES Duesseldorf GmbH
European Research Center
Riesstrasse 25, 80992 München
E-mail: radu.tudo...@huawei.com
Mobile: +49 15209084330 <+49%201520%209084330>
Telephone: +49 891588344173 <+49%2089%201588344173>
HUAWEI TECHNOLOGIES Duesseldorf GmbH
Hansaallee 205, 40549 Düsseldorf, Germany, www.huawei.com
Registered Office: Düsseldorf, Register Court Düsseldorf, HRB
56063,
Managing Director: Bo PENG, Wanzhou MENG, Lifang CHEN
Sitz der Gesellschaft: Düsseldorf, Amtsgericht Düsseldorf, HRB
56063,
Geschäftsführer: Bo PENG, Wanzhou MENG, Lifang CHEN
This e-mail and its attachments contain confidential information
from
HUAWEI, which is intended only for the person or entity whose
address
is
listed above. Any use of the information contained herein in any
way
(including, but not limited to, total or partial disclosure,
reproduction,
or dissemination) by persons other than the intended recipient(s)
is
prohibited. If you receive this e-mail in error, please notify
the
sender
by phone or email immediately and delete it!
-----Original Message-----
From: Fabian Hueske [mailto:fhue...@gmail.com]
Sent: Tuesday, February 14, 2017 1:01 AM
To: dev@flink.apache.org
Subject: [DISCUSS] Table API / SQL indicators for event and
processing
time
Hi,
I'd like to start an discussion about how Table API / SQL queries
indicate
whether an operation is done in event or processing time.
1) Why do we need to indicate the time mode?
We need to distinguish event time and processing time mode for
operations
in queries in order to have the semantics of a query fully
defined.
This cannot be globally done in the TableEnvironment because some
queries
explicitly request an expression such as the ORDER BY clause of
an
OVER
window with PRECEDING / FOLLOWING clauses.
So we need a way to specify something like the following query:
SELECT
a,
SUM(b) OVER (PARTITION BY c ORDER BY proctime ROWS BETWEEN 2
PRECEDING
AND CURRENT ROW) AS sumB, FROM myStream
where "proctime" indicates processing time. Equivalently
"rowtime"
would
indicate event time.
2) Current state
The current master branch implements time support only for
grouping
windows in the Table API.
Internally, the Table API converts a 'rowtime symbol (which looks
like
a
regular attribute) into a special expression which indicates
event-time.
For example:
table
.window(Tumble over 5.milli on 'rowtime as 'w)
.groupBy('a, 'w)
.select(...)
defines a tumbling event-time window.
Processing-time is indicated by omitting a time attribute
(table.window(Tumble over 5.milli as 'w) ).
3) How can we do that in SQL?
In SQL we cannot add special expressions without touching the
parser
which
we don't want to do because we want to stick to the SQL standard.
Therefore, I see only two options: adding system attributes or
(parameterless) built-in functions. I list some pros and cons of
the
approaches below:
1. System Attributes:
+ most natural way to access a property of a record.
+ works with joins, because time attributes can be related to
tables
- We need to ensure the attributes are not writable and always
present
in
streaming tables (i.e., they should be system defined
attributes).
- Need to adapt existing Table API expressions (will not change
the
API
but some parts of the internal translation)
- Event time value must be set when the stream is converted,
processing
time is evaluated on the fly
2. Built-in Functions
+ Users could try to modify time attributes which is not possible
with
functions
- do not work with joins, because we need to address different
relations
- not a natural way to access a property of a record
I think the only viable choice are system attributes, because
built-in
functions cannot be used for joins.
However, system attributes are the more complex solution because
they
need
a better integration with Calcite's SQL validator (preventing
user
attributes which are named rowtime for instance).
Since there are currently a several contributions on the way
(such
as
SQL
OVER windows FLINK-5653 to FLINK-5658) that need time indicators,
we
need a
solution soon to be able to make progress.
There are two PRs, #3252 and #3271, which implement the built-in
marker
functions proctime() and rowtime() and which could serve as a
temporary
solution (since we do not work on joins yet).
I would like to suggest to use these functions as a starting
point
(once
the PRs are merged) and later change to the system attribute
solution
which
needs a bit more time to be implemented.
I talked with Timo today about this issue and he said he would
like
to
investigate how we can implement this as system functions
properly
integrated with Calcite and the SQL Validator.
What do others think?
Best, Fabian
