Hello,
there were old and recent discussions about the representation of time in accounting records. At this moment the acc module stores the unix timestamp as datetime value.
In some countries is required to store more accuracy, beyond the seconds. Also, the datetime gives some troubles with converting back to unix timestamp, specifically with time zones and daylight saving times.
I want to open a discussion that is visible for all users and devs, being something affecting probably everyone, input from anyone that is interested being relevant to select the right approach.
Here are some suggestions presented so far.
1) store seconds.miliseconds as double - there is a patch (which probably needs some tunings itself) on tracker. The precision is shifted from seconds to milliseconds. Internally computed from microseconds from timeval structure. The other modules might need updates (iirc dbtext has only 2 decimals precisions for double, while miliseconds will require at least three -- could be the case for flatstore or other modules).
Advantage is the computing of duration directly by subtraction of two values.
2) store seconds and microseconds as two separate values. Should be no issues with existing modules. Even more accuracy than above, no issues with other modules, but will require to use two columns (thus four values to compute the duration)
Any other suggestions?
Personally, at this moment, I will go with 2), but that might not meet everyone's needs. So a solution can be to make it configurable, as a bitmask of what time representation to store.
Say, there will be a new parameter timestamp_mode for acc module: - bit one set - store seconds timestamp as for now (default) - bit two set - store seconds and microseconds in separate integer columns - bit three set - store seconds.miliseconds as double value in one column
I could add the parameter and tune acc for bit one and two. For bit three a deeper review is needed for other modules, probably the developer that submitted to the tracker can do it.
More suggestions? Pro or cons opinions?
Cheers, Daniel
On Monday 29 April 2013 11:05:36 Daniel-Constantin Mierla wrote:
Here are some suggestions presented so far.
- store seconds.miliseconds as double - there is a patch (which
Please do not use floating point respresentations for values that will be used in accounting. Floating point is imprecise. As the time related columns will most probably be used for billing, the values should be exact. In SQL this means using the DECIMAL or NUMERIC column type.
- store seconds and microseconds as two separate values. Should be no
issues with existing modules. Even more accuracy than above, no issues with other modules, but will require to use two columns (thus four values to compute the duration)
Difficult to use in calculations.
Any other suggestions?
3) Use native mili/microseconds support for DATETIME or TIMESTAMP in the database. At least MariaBD and PostgreSQL support this.
4) Store mili/microseconds since epoch in a BIGINT column.
Say, there will be a new parameter timestamp_mode for acc module:
- bit one set - store seconds timestamp as for now (default)
- bit two set - store seconds and microseconds in separate integer columns
- bit three set - store seconds.miliseconds as double value in one column
- bit three set - store seconds.miliseconds as DECIMAL value in one column - bit four set - add mili/microseconds to DATETIME (only valid when bit 1 is set too) - bit five set - store mili/microseconds since epoch as BIGINT value in one column
Alternatively, 2 settings can be used, one for storage format and one to choose the precision/resolution. This provides the most flexibility for the user.
timestamp_format: datetime (TIMESTAMP or DATETIME) epoch ((BIG)INT or DECIMAL, depending on resolution) split_epoch (2x INT)
timestamp_resolution: seconds, miliseconds, microseconds
Default would be the current situation: datetime + seconds
Hello,
On 4/29/13 1:42 PM, Alex Hermann wrote:
On Monday 29 April 2013 11:05:36 Daniel-Constantin Mierla wrote:
Here are some suggestions presented so far.
- store seconds.miliseconds as double - there is a patch (which
Please do not use floating point respresentations for values that will be used in accounting. Floating point is imprecise. As the time related columns will most probably be used for billing, the values should be exact. In SQL this means using the DECIMAL or NUMERIC column type.
There are non-sql database connectors, so it has to be something working for everything. That's why I proposed variants, even possible to store couple of formats at the same time.
- store seconds and microseconds as two separate values. Should be no
issues with existing modules. Even more accuracy than above, no issues with other modules, but will require to use two columns (thus four values to compute the duration)
Difficult to use in calculations.
Probably easier that dealing with formatting back from datetime and take care of timezone and daylight saving time.
Any other suggestions?
- Use native mili/microseconds support for DATETIME or TIMESTAMP in the
database. At least MariaBD and PostgreSQL support this.
There are other db_* connectors that have to handle nicely the new addition.
- Store mili/microseconds since epoch in a BIGINT column.
Can be added if anyone writes the code for the option and checks/updates the db connectors for compatibility.
Say, there will be a new parameter timestamp_mode for acc module:
- bit one set - store seconds timestamp as for now (default)
- bit two set - store seconds and microseconds in separate integer columns
- bit three set - store seconds.miliseconds as double value in one column
- bit three set - store seconds.miliseconds as DECIMAL value in one column
- bit four set - add mili/microseconds to DATETIME (only valid when bit 1 is
set too)
- bit five set - store mili/microseconds since epoch as BIGINT value in one
column
Alternatively, 2 settings can be used, one for storage format and one to choose the precision/resolution. This provides the most flexibility for the user.
timestamp_format: datetime (TIMESTAMP or DATETIME) epoch ((BIG)INT or DECIMAL, depending on resolution) split_epoch (2x INT)
timestamp_resolution: seconds, miliseconds, microseconds
Default would be the current situation: datetime + seconds
At this moment is datetime.
Cheers, Daniel
-
Alex Hermann -
Daniel-Constantin Mierla