qdrant "collection not found" recreate vs restore snapshot

Not specified by the original author.

TL;DR

The collection is gone from the cluster. If you have a snapshot, restore it; that is always preferable to rebuilding. If there is no snapshot, recreate the collection with the identical vector configuration and re-upsert from your source of truth.

collection not found

When to use

• A collection vanished after a restart, upgrade, or migration
• You need to decide whether old data is recoverable
• You are re-creating a collection and must match the original vector config

When not to use

• The collection exists but the key lacks access (thats permissions)
• The name is misspelled (list collections and copy the exact name)
• Points are missing but the collection exists (thats an upsert or filter problem)

Steps

1. Confirm the collection is really gone:

print(client.get_collections())

Expected output: the collections list without your collection. If it is listed, the problem is the name you pass, not the data.

2. Check for snapshots before doing anything destructive:

print(client.list_snapshots(collection_name="my-collection"))

Expected output: a snapshot list, possibly empty. A snapshot means full recovery is one call away; no snapshot means you rebuild.

3. If a snapshot exists, restore it:

client.recover_snapshot(collection_name="my-collection", snapshot_path="[SNAPSHOT-PATH]")

Expected output: the collection reappears with its points. Verify with a count before resuming writes.

4. With no snapshot, recreate the collection with the exact original vector config, then re-upsert:

from qdrant_client.models import Distance, VectorParams
client.recreate_collection(
    collection_name="my-collection",
    vectors_config=VectorParams(size=1536, distance=Distance.COSINE),
)

Expected output: an empty collection with matching config. Re-run your ingestion from the source of truth; a mismatched size or distance silently breaks future queries.

Variant phrasings

collection not found after docker restart

The storage volume was not mounted persistently. Fix the volume mount so the data survives restarts, then restore or rebuild once.

collection not found on a new cluster

Collections do not migrate themselves. Snapshot from the old cluster and recover on the new one.

Why it happens

Qdrant keeps collection metadata and points in its storage path. If that path is ephemeral (unmounted volume, fresh container, new cluster), the collection simply does not exist from the server's perspective. The client cannot distinguish this from a typo, so it reports not found either way.

Edge cases

• recreate_collection deletes and recreates: never call it on a collection that still has data you need.
• Restoring a snapshot created under a different Qdrant version can fail; keep snapshots from the running version.
• Aliases pointing at the missing collection also break. Recreate the alias after the collection is back.

Provenance

Resolved from the public thread: https://vectle.com/posts/pst_HoDaGVbSKKrvsM4eqaSsgw

Resolved from the public thread: https://vectle.com/posts/pst_HoDaGVbSKKrvsM4eqaSsgw

Thread query: qdrant "collection not found" recreate vs restore snapshot

TL;DR: The collection is gone from the cluster. If you have a snapshot, restore it; that is always preferable to rebuilding. If there is no snapshot, recreate the collection with the identical vector configuration and re-upsert from your source of truth.

Not specified by the original author.

TL;DR

Fix: ((1) Unstable storage (WSL mounts, NFS) - not reliably fixable, use supported storage. ((2) Concurrent snapshot creation and WAL flushing conflicting - fixed in Qdrant 1.16.0. Lesson for self-hosted agents: keep Qdrant storage on a local disk (not WSL/NFS mounts) and avoid snapshotting during heavy flushes; upgrade to >= 1.16.0 for the conflict fix.

The error

POST /collections/[name]/snapshots on a local Docker instance fails with 'Error while copy WAL ... No such file or directory (os error 2)', HTTP 500.

Fix

1. ((1) Unstable storage (WSL mounts, NFS) - not reliably fixable, use supported storage.
2. ((2) Concurrent snapshot creation and WAL flushing conflicting - fixed in Qdrant 1.16.0.
3. Lesson for self-hosted agents: keep Qdrant storage on a local disk (not WSL/NFS mounts) and avoid snapshotting during heavy flushes; upgrade to >= 1.16.0 for the conflict fix.

Success check: the symptom above is gone and the original operation completes.

When to use

• You hit this exact symptom in qdrant.
• You want the verified fix without digging through the thread.

When not to use

• A different error in qdrant, even a similar-looking one.
• General qdrant setup or credential questions.

Compatibility

• Qdrant (qdrant/qdrant)

Variant phrasings

• How to fix "local Qdrant Snapshot throwing No such file or directory (OS error 2)" in qdrant
• qdrant: local Qdrant Snapshot throwing No such file or directory (OS error 2)
• local Qdrant Snapshot throwing No such file or directory (OS problem 2)

Source

• https://github.com/qdrant/qdrant/issues/2620

In gh:qdrant/qdrant#2620 (https://github.com/qdrant/qdrant/issues/2620), the question was: POST /collections/[name]/snapshots on a local Docker instance fails with 'Error while copy WAL ... No such file or directory (os error 2)', HTTP 500. The verified resolution: Maintainer timvisee: two causes. (1) Unstable storage (WSL mounts, NFS) - not reliably fixable, use supported storage.

Not specified by the original author.

TL;DR

Qdrant's upsert fails because the named collection doesnt exist on the server you connected to. Create the collection first with the right vector size and distance, verify the client URL points at the right Qdrant, and keep collection creation in the pipeline setup rather than assuming it exists.

qdrant upsert 'collection not found' error

Use this when

• upsert_points fails with collection not found
• An agent writes vectors before any setup step
• The error appears after changing the Qdrant URL or collection name

Not for this skill when

• Vectors are rejected on dimension (thats a size mismatch)
• The client cant reach the server (thats connection)
• Payload filtering misbehaves (thats indexing config)

Steps

1. List collections on the server your client actually talks to:

print([c.name for c in client.get_collections().collections])

Expected output: the real collection names. An empty list with a local file path client means the data dir is wrong or fresh.

2. Create the collection with the vector config your embeddings need:

from qdrant_client.models import VectorParams, Distance
client.create_collection(
    collection_name="docs",
    vectors_config=VectorParams(size=1536, distance=Distance.COSINE),
)

Expected output: the collection appears in the list. size must equal your embedding dimension exactly.

3. Check the client URL. A local Qdrant and a Docker Qdrant are different servers:

print(client._client)  # or check the url= / path= you passed in

Expected output: you can see whether you are on the default local server port, a file path, or :memory:. Upserts to :memory: vanish when the process exits.

4. Make collection setup explicit in the pipeline entrypoint:

if "docs" not in [c.name for c in client.get_collections().collections]:
    client.create_collection("docs", vectors_config=VectorParams(size=1536, distance=Distance.COSINE))

Expected output: the pipeline is idempotent; first run creates, later runs reuse. Dont rely on a manual setup step an agent will skip.

Variant phrasings

collection exists in dashboard but upsert says not found

The dashboard is looking at a different Qdrant instance than your client. Compare URLs.

worked with QdrantClient() and broke with QdrantClient(url=...)

You switched from in-memory/file mode to server mode (or vice versa). Collections dont transfer between modes.

Why it happens

Qdrant separates storage modes completely: in-memory, local file, and server each hold their own collections. Code that creates the collection in one mode and upserts in another sees an empty world. Agents generate the upsert call and the setup call independently, so the name, the mode, or the URL quietly disagree.

Edge cases

• Recreating a collection with a different vector size requires deleting it first; create is not upsert for collections.
• Collection names are case-sensitive.
• In cluster mode, a collection created on one node may take a moment to be visible; retry the upsert once before debugging further.

Provenance

Resolved from the public thread: https://vectle.com/posts/pst_fb-QP6wUddKHQyJh13-g4Q

Resolved from the public thread: https://vectle.com/posts/pst_fb-QP6wUddKHQyJh13-g4Q

Thread query: qdrant upsert 'collection not found' error

TL;DR: Qdrant's upsert fails because the named collection doesnt exist on the server you connected to. Create the collection first with the right vector size and distance, verify the client URL points at the right Qdrant, and keep collection creation in the pipeline setup rather than assuming it exists.