Correcting insertion of new data

[juvenalb]

I have a fairly simple schema:

CREATE TABLE "mysensor" (
"id" INTEGER PRIMARY KEY AUTOINCREMENT NOT NULL,
"date" INTEGER NOT NULL,
"sensor_data" INTEGER NOT NULL
);

When inserting data into this table, the 3rd party application takes care of the date field, it's a standard UNIX-timestamp in seconds.

Until now this was acceptable, as the sensors I used sent their data:

• either at low frequency (1 sample/minute, or even 1 sample/day)
• or they were not too accurate, so the measured data had to change a lot to trigger the insertion (there is a constraint in the 3rd party app that it doesn't insert if last value matches the current one)

However I got some new sensors which are more precise, so the triggering in the app can't filter as the measured data keeps changing a little so technically last measured is not equal to current one, so the insertion will take place.

So, anyway I would like to solve this from DB side, without too much overhead.

I know how I can silently abort the insertion with TRIGGER, however I'm looking for an alternative best-effort method to have update the table when a too fresh data arrives.

Too fresh: let's consider that everything below 1 minute.

So currently, I'm dropping all data within one minute:

CREATE TRIGGER "mytrigger" BEFORE INSERT ON "mysensor" WHEN NEW.date < 60 + (SELECT MAX(date) FROM "mysensor")
BEGIN
  SELECT RAISE(ABORT, "Last data is yet too fresh");
END

I thought on some wise averaging, but for correct averaging I would need an extra count variable too.

So currently I can either

• lose data correctness by incorrectly weighting earlier vs new data, or
• lose storage by adding a new field to schema to store the count of averaged earlier data.

Is there a better way to keep data-correctness while not sacrificing storage?

[guilherme]

I'm not sure if it's wise to drop data. Maybe your current assumption changes in the future, or perhaps you'll need the extra information later.

If possible, I would do the following:

1 - Create a new table that collects all raw data

CREATE TABLE "sensorRawData" (
"id" INTEGER PRIMARY KEY AUTOINCREMENT NOT NULL,
"date" INTEGER NOT NULL,
"sensor_data" INTEGER NOT NULL,
"sqltime" TIMESTAMP DEFAULT CURRENT_TIMESTAMP NOT NULL
);

2 - Only insert the data you need from sensorRawData into your mysensor table (either as a trigger or a recurring job), by calculating the difference between "date" and "sqltime".

3 - If space is a limitation, run a weekly cleanup job that deletes data older than X weeks or X months, depending on your configuration.