[0.9.3] DB should not crash when using invalid expression "GROUP BY time"

[jeremyVignelles]

Hi,

From a fresh database, i created a point with the admin by writing:

test value=0

Then, issuing the query

SELECT value FROM test GROUP BY time

crashes the database.
Here is the log : http://pastebin.com/g6DqnF43

By the way, is there any way I can select all the timestamps for a measurement (optionally filtered by a where clause) ?

Thanks for your work

[desa]

@jeremyVignelles thank you for reporting this. I'm getting the same error.

panic: interface conversion: interface is float64, not map[string]interface {}

goroutine 128 [running]:
github.com/influxdb/influxdb/tsdb.(*limitedRowWriter).processValues(0xc2080f0600, 0xc208a04000, 0x2, 0x2710, 0x0)
    /Users/michaeldesa/go/src/github.com/influxdb/influxdb/tsdb/executor.go:741 +0x1083
github.com/influxdb/influxdb/tsdb.(*limitedRowWriter).Flush(0xc2080f0600)
    /Users/michaeldesa/go/src/github.com/influxdb/influxdb/tsdb/executor.go:667 +0xd6
github.com/influxdb/influxdb/tsdb.(*SelectExecutor).executeRaw(0xc2088cd1a0, 0xc20898f0e0)
    /Users/michaeldesa/go/src/github.com/influxdb/influxdb/tsdb/executor.go:234 +0x4e2
created by github.com/influxdb/influxdb/tsdb.(*SelectExecutor).Execute
    /Users/michaeldesa/go/src/github.com/influxdb/influxdb/tsdb/executor.go:94 +0x9a

[desa]

@jeremyVignelles in regards to By the way, is there any way I can select all the timestamps for a measurement (optionally filtered by a where clause) ? can you be a little more specific? An example of the functionality you're looking for would be good.

[jeremyVignelles]

@mjdesa : I meant something like this:

SELECT time FROM measurement WHERE time > now() - 3d GROUP BY time

The idea behind this is to provide a timeline for events that happens unfrequently (once a day for example), where the user can click to view the data (128 fields * 768 rows with the same timestamp).
Is there a way to achieve this?

[beckettsean]

@jeremyVignelles I don't understand what would be different about the output from

SELECT time FROM measurement WHERE time > now() - 3d GROUP BY time
vs
SELECT * FROM measurement WHERE time > now() - 3d

Can you maybe give some example output showing what you are describing?

[jeremyVignelles]

In the first case, you would get only the distinct timestamps, whereas the second would send all the data with the request.
The user would click on a time point afterwards, and a query like "SELECT * FROM test WHERE time = timestamp" would retrieve the actual data.
I mentioned 3days in my example query, but the timeline I would like to implement should be able to show the time points for the last month.
The timeline itself would be displayed in a browser with some js, and I don't want the user's memory to be overloaded with useless data.

[zimbatm]

@jeremyVignelles if your data is sparse it's probably better to return all the data points instead of having most timestamps filled with null. GROUP BY time(1s) is probably good enough otherwise.

[jeremyVignelles]

@zimbatm This is what I want, select all the timestamp for all the datapoints, but I would like to remove timestamp duplicates for that, and I would expect the expression GROUP BY time to act this way:

• Take all the data points
• Create groups of data points which have the exact same timestamp
• Return one row per group, with the given timestamp (very quickly, of course 😉 )

Indeed, I do not want extra null values which will be polluting the data when using GROUP BY time(1s). Most of the time, I will have one timestamp per day, but when doing manual tests, there could be 2 or 3 timestamps in 15 minutes...

[beckettsean]

@jeremyVignelles It sounds like you want something more or less like SELECT DISTINCT(time) FROM ... to get a list of all the times at which one ore more points exist.

Please open a feature request in a new issue for that, as this one is now about preventing the panic from the incorrect query.