ES|QL: improper pruning of unused columns at planning time #126017
Description
With the CSV dataset
from sample_data
| eval type = 1000
| lookup join message_types_lookup on message
| grok type "%{WORD:foo}"
| keep event_duration, type
{
"took": 39,
"is_partial": false,
"columns": [
{
"name": "event_duration",
"type": "long"
},
{
"name": "type",
"type": "keyword"
}
],
"values": [
....
]
type (a keyword) comes from the right-hand side of the JOIN, and overwrites the previous integer type column.
Removing it from the final projections also removes it from the output of the JOIN, no longer masking the existing integer column:
from sample_data
| eval type = 1000
| lookup join message_types_lookup on message
| grok type "%{WORD:foo}"
| keep event_duration
{
"error": {
"root_cause": [
{
"type": "verification_exception",
"reason": "Found 1 problem\nline 1:91: Grok only supports KEYWORD or TEXT values, found expression [type] type [INTEGER]"
}
],
"type": "verification_exception",
"reason": "Found 1 problem\nline 1:91: Grok only supports KEYWORD or TEXT values, found expression [type] type [INTEGER]"
},
"status": 400
}