I’m getting occasional 409 errors with the

ext-firestore-typesense-search-indexToTypesenseOnFirestoreWrite

cloud function. I don’t understand why this would throw an error as it shouldn’t matter if a document exists to update it… any idea what I’m doing wrong here?

Could you post the full log message with a few lines before and after?

Sorry this is a few months behind... Here's a recent instance of the error. Log says "creating document" and not the usual "upserting" which I'm sure is part of this problem but its unclear how. This is the log from document creation:

{
  "textPayload": "Creating document {\"measure_text_length\":343,\"rank\":0.5,\"text\":\"Indexers certainly have a very strong want for data determinism, they are constantly monitoring and worried about their data veracity. it's very time consuming\\nConsumers are also feeling this too since we've seen cases with \\\"wrong\\\" data returned by queries\\nBut no we have not had an attack like that yet that we know of, but what's your point?\",\"title\":\"slack\",\"type\":\"message\",\"url\":\"<https://blockscienceteam.slack.com/archives/C036NAWP0CT/p1677261752395249?cid=C036NAWP0CT&thread_ts=1676059661.243089>\",\"id\":\"aHR0cHM6Ly9ibG9ja3NjaWVuY2V0ZWFtLnNsYWNrLmNvbS9hcmNoaXZlcy9DMDM2TkFXUDBDVC9wMTY3NzI2MTc1MjM5NTI0OT9jaWQ9QzAzNk5BV1AwQ1QmdGhyZWFkX3RzPTE2NzYwNTk2NjEuMjQzMDg5\"}",
  "insertId": "63f8fbbb0005cfd5c189c11f",
  "resource": {
    "type": "cloud_function",
    "labels": {
      "project_id": "knowledge-management-333914",
      "region": "us-central1",
      "function_name": "ext-firestore-typesense-search-indexToTypesenseOnFirestoreWrite"
    }
  },
  "timestamp": "2023-02-24T18:02:35.380885Z",
  "severity": "DEBUG",
  "labels": {
    "instance_id": "00c61b117c5a189878a95fae39681518ed338d4fa826a35986e68469d2ec5b3070fede155da6333715bb7dfc26d0c4c0f650d7a4a775945d9115",
    "execution_id": "qu4q7bkf0lq7"
  },
  "logName": "projects/knowledge-management-333914/logs/cloudfunctions.googleapis.com%2Fcloud-functions",
  "trace": "projects/knowledge-management-333914/traces/a9c2419e85dd694ae3fe11d14d31bebf",
  "receiveTimestamp": "2023-02-24T18:02:35.518729400Z"
}

And this is the error:

{
  "textPayload": "Error: Request failed with HTTP code 409 | Server said: A document with id aHR0cHM6Ly9ibG9ja3NjaWVuY2V0ZWFtLnNsYWNrLmNvbS9hcmNoaXZlcy9DMDM2TkFXUDBDVC9wMTY3NzI2MTc1MjM5NTI0OT9jaWQ9QzAzNk5BV1AwQ1QmdGhyZWFkX3RzPTE2NzYwNTk2NjEuMjQzMDg5 already exists.\n    at ApiCall.customErrorForResponse (/workspace/node_modules/typesense/lib/Typesense/ApiCall.js:229:21)\n    at ApiCall.performRequest (/workspace/node_modules/typesense/lib/Typesense/ApiCall.js:118:48)\n    at processTicksAndRejections (internal/process/task_queues.js:95:5)",
  "insertId": "63f8fbbb000bdb5feae60206",
  "resource": {
    "type": "cloud_function",
    "labels": {
      "project_id": "knowledge-management-333914",
      "region": "us-central1",
      "function_name": "ext-firestore-typesense-search-indexToTypesenseOnFirestoreWrite"
    }
  },
  "timestamp": "2023-02-24T18:02:35.777055Z",
  "severity": "ERROR",
  "labels": {
    "instance_id": "00c61b117c5a189878a95fae39681518ed338d4fa826a35986e68469d2ec5b3070fede155da6333715bb7dfc26d0c4c0f650d7a4a775945d9115",
    "execution_id": "qu4q7bkf0lq7"
  },
  "logName": "projects/knowledge-management-333914/logs/cloudfunctions.googleapis.com%2Fcloud-functions",
  "trace": "projects/knowledge-management-333914/traces/a9c2419e85dd694ae3fe11d14d31bebf",
  "receiveTimestamp": "2023-02-24T18:02:35.852702927Z"
}

Hmm, so it looks like this document already exists in Typesense, but Firestore is calling the extension as a new document…

In the above case I see a log for upserting and then creating the doc, in other cases where creation precedes updating it seems to run fine

I see… This is most probably a result of Firestore not triggering change events in order unfortunately: https://github.com/typesense/firestore-typesense-search/issues/32

Yeah I wondered about that from the seemingly random probability of this error

Question for you: I was thinking of this as a solution:

One potential solution to this could be to query Firestore on each change trigger and push the latest version of the Firestore document to Typesense, instead of using the snapshot document from the event

But from a cost perspective, do you think doing these reads for each Firestore write is reasonable? I guess it will essentially increase write costs by 33%?

Hmmmm, hard to say, it feels very much like a workaround. It would probably work for my case but only because it's not already a high-cost function.

Yeah that would solve the 409 error from showing up… But then if the create shows up after the update, the document will now be stale in Typesense, since the create event will have a stale version of the doc (when it was created) in its payload

Wouldn't it just mean you'd have two update_or_create calls? So the order of events in my case would be a bit like: 1. document added to firestore 2. TS update_or_create triggered -> TS document doesn't exist, creates it 3. document updated with "platform" field 4. TS update_or_create triggered -> TS document exists, updates it Ahhh... I see what you mean now...

Right

The other solution mentioned on the GH issue, does that sync state need to be stored in firestore? if write triggers have an accurate timestamp can that be used to order TS updates?

We still need to store the state of what the last synced timestamp is for each document, to be able to know whether to discard an event

To avoid redundant calls to typesense?

To prevent processing stale events

Right

Unfortunately yeah… That’s why I didn’t make this change. Figured if this is an issue for a particular use case, the 409 error will at least be a hint that this is happening

So the options are: • pull from firestore directly instead of using the snapshot -> increase write costs • track sync state in a collection -> increased storage usage (and still need more than just the snapshot anyway? I guess this requires the first option too) • periodically sync -> increased cost from queries and storage and unneeded periodic checks when no activity is happening?

I think #1 is the simplest to expose as a config option… #2 also increases storage costs, so #1 might be better #3 requires a completely different extension, since it doesn’t rely on triggers

Yeah and I suppose #3 cant be automatically configured as an extension afaik and would require extra action to setup periodic events/triggers in GCP?

Correct

Gotcha, I would definitely turn on a config option for #1 if it was there. Not because it's vitally important to be in perfect sync for users but because it makes debugging so much easier when the Firestore-TS link is 100% predictable. Then I know a bug has to be my fault 🤣

Haha!

Yeah sure, if I get a chance I can try and create a PR, seems like it should be a simple change looking at the src now

A PR would be awesome!

Got any links to help test firestore extensions? Or I can just PR and cross my fingers (or you could test it on your end)

These instructions should work: https://github.com/typesense/firestore-typesense-search#development-workflow

Do you think the flag should be "read from firestore" (off by default) or the opposite, "read from event" (on by default)?

Thank you! I’ll take a look this coming week or early week after.