Rows written with `INSERT OR IGNORE` still read/writing

[hbmartin]

I am frequently inserting datasets with many duplicates using INSERT OR IGNORE .... However, it appears that this still causes writes (and also maybe reads?) on the remote replicant? Is this expected behavior? Is there a configuration to prevent this?

[sivukhin]

Hi @hbmartin!

I think that reads are expected for INSERT OR IGNORE query but there must be no writes actually.

Can you share more details about your setup?

• "it appears that this still causes writes" - why do you think so? Do you see rows_written stats in the Hrana response?
• What version of LibSQL do you use?
• How do you use LibSQL? Do you query it from sqld with Hrana protocol?
• If you can share some example which will reproduce the problem - this will be great!

BTW, I tried to replicate your issue but don't see any signs of writes for your type of query:

$> cat query.json
{
  "requests": [
    {
      "type": "execute", "stmt": { "sql": "INSERT OR IGNORE INTO with_dup VALUES (1, 10);" }
    },
    {
      "type": "close"
    }
  ]
}
$> cat query.json | curl -X POST https://localhost:8080/v2/pipeline -d @- | jq . # some fields are omitted
{
  "results": [
    {
      "type": "ok",
      "response": {
        "type": "execute",
        "result": {
          "rows_read": 1,
          "rows_written": 0,
        }
      }
    },
    {
      "type": "ok",
      "response": {
        "type": "close"
      }
    }
  ]
}