gap filling: fill previous?

[est]

Taken from the gap filling doc

    date    | volume
------------+--------
 2017-09-30 |      0
 2017-09-29 |  11315
 2017-09-28 |   8216
 2017-09-27 |   5591
 2017-09-26 |   9182
 2017-09-25 |  14359
 2017-09-24 |      0
 2017-09-23 |      0
 2017-09-22 |   9855

http://docs.timescale.com/v0.9/using-timescaledb/reading-data#gap-filling

Now I dont want to fill(0), I need fill(previous) like this:

    date    | volume
------------+--------
 2017-09-30 |      0
 2017-09-29 |  11315
 2017-09-28 |   8216
 2017-09-27 |   5591
 2017-09-26 |   9182
 2017-09-25 |  14359
 2017-09-24 |  14359
 2017-09-23 |  14359
 2017-09-22 |   9855

How can this be done?

[est]

I found this clever windowing function hack

https://stackoverflow.com/a/19012333/41948

[cevian]

We plan to improve this functionality shortly, but for now you can use the method here: https://www.joyofdata.de/blog/locf-linear-imputation-postgresql-tutorial/

[mfreed]

@est This capability can now be found using our time_bucket_gapfill and locf functionality, first released in version 1.2.

https://docs.timescale.com/v1.2/api#locf

[est]

@mfreed thanks!

[slidenerd]

• Instead of using any form of averaging or previous value substitution, I am querying the external API to update the missing data by detecting gaps with a task queue

• If I had say 1 million 1m data points and I wanted 5m data points from it, do I have to run over the whole database again or just the latest chunk

• Are the older chunks updatable to fill missing data because it may take some time with a task queue