Consider enforcing no leading underscore in tag keys

[e-dard]

I would like to consider ensuring that the storage engine does not accept any tag keys with leading underscores, essentially leaving the _ prefix as a private key space for internal series key information.

With that as an invariant, we could make certain performance improvements around reading and parsing keys.

[e-dard]

This was resolved via #12391

[zhulongcheng]

measurement overwrite _measurement tag in flux table.

// write
any,_field=ff,_measurement=mm,_tag=tt foo="bar"

// query: "from(bucket:"purple")|> range(start:-10h)"
,result,table,_start,_stop,_time,_value,_field,_measurement,_tag
,_result,1,2019-05-31T08:09:08.867394Z,2019-05-31T08:19:08.867394Z,2019-05-31T08:19:00Z,bar,foo,any,tt

at least, _start,_stop,_time,_value,_field,_measurement these tag keys should not be accepted.

[zhulongcheng]

influxdb/storage/readservice/cursor.go

Lines 143 to 153 in 530ceb9

	// Normalise the special tag keys to the emitted format.
	mv := c.row.Tags.Get(models.MeasurementTagKeyBytes)
	c.row.Tags.Delete(models.MeasurementTagKeyBytes)
	c.row.Tags.Set(measurementKeyBytes, mv)
	c.row.Tags.Delete(models.FieldKeyTagKeyBytes)
	c.row.Tags.Set(fieldKeyBytes, fv)
	return &c.row
	}

[e-dard]

@zhulongcheng thanks for the issue! Hmm I think this is an issue in the read side. @stuartcarnie can you take a look?

[stuartcarnie]

@e-dard the issue I see here is that _measurement was allowed via a write, and therefore stored as a real tag. Specifically, the series key is stored as:

<org><bucket>,\x00=any,_field=ff,_measurement=mm,_tag=tt,\xff=foo

which should never be allowed.

When a read occurs, the \x00 and \xff bytes are remapped to _measurement and _field respectively, overwriting those stored tags:

influxdb/storage/readservice/cursor.go

Lines 143 to 153 in 530ceb9

	// Normalise the special tag keys to the emitted format.
	mv := c.row.Tags.Get(models.MeasurementTagKeyBytes)
	c.row.Tags.Delete(models.MeasurementTagKeyBytes)
	c.row.Tags.Set(measurementKeyBytes, mv)
	c.row.Tags.Delete(models.FieldKeyTagKeyBytes)
	c.row.Tags.Set(fieldKeyBytes, fv)
	return &c.row
	}

[e-dard]

@stuartcarnie yeah seems about right. I think then we’ll have to check and block these. I’ll put this on the back log.

[stale]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.

[zhulongcheng]

don't close.

[8none1]

Could this be done at the API layer? Would internal calls use the same APIs and so be rejected?