Switched ML Model in web interface, now immich-server keeps crashing
#help-desk-support
I switched to this new model and now the clip_index is wrong, but clip_index doesn't exist?
I tried just deleting the ML cache, no luck, same issue. There is no env variable to set the ML Model? I was hoping I could just set it back now.
Any help would be greatly appreciated!
[Nest] 7 - 04/26/2025, 10:39:44 PM LOG [Microservices:SmartInfoService] Dimension size of model immich-app/ViT-SO400M-16-SigLIP2-384__webli is 1152, but database expects 1024.
[Nest] 7 - 04/26/2025, 10:39:44 PM LOG [Microservices:SmartInfoService] Updating database CLIP dimension size to 1152.
Query failed : {
durationMs: 1.060765000000174,
error: PostgresError: relation "clip_index" does not exist
at ErrorResponse (/usr/src/app/node_modules/postgres/cjs/src/connection.js:790:26)
at handle (/usr/src/app/node_modules/postgres/cjs/src/connection.js:476:6)
at Socket.data (/usr/src/app/node_modules/postgres/cjs/src/connection.js:315:9)
at Socket.emit (node:events:518:28)
at addChunk (node:internal/streams/readable:561:12)
at readableAddChunkPushByteMode (node:internal/streams/readable:512:3)
at Readable.push (node:internal/streams/readable:392:5)
at TCP.onStreamRead (node:internal/stream_base_commons:189:23) {
severity_local: 'ERROR',
severity: 'ERROR',
code: '42P01',
file: 'namespace.c',
line: '433',
routine: 'RangeVarGetRelidExtended'
},
sql: 'reindex index clip_index',
params: []
}
5 Replies
Here is my postgres log:
2025-04-26 23:07:16.239 UTC [1] LOG: starting PostgreSQL 14.10 (Debian 14.10-1.pgdg120+1) on x86_64-pc-linux-gnu, compiled by gcc (Debian 12.2.0-14) 12.2.0, 64-bit
2025-04-26 23:07:16.239 UTC [1] LOG: listening on IPv4 address "0.0.0.0", port 5432
2025-04-26 23:07:16.239 UTC [1] LOG: listening on IPv6 address "::", port 5432
2025-04-26 23:07:16.263 UTC [1] LOG: listening on Unix socket "/var/run/postgresql/.s.PGSQL.5432"
2025-04-26 23:07:16.287 UTC [28] LOG: database system was shut down at 2025-04-26 23:07:06 UTC
[2025-04-26T23:07:16Z INFO service::utils::clean] Find directory "pg_vectors/indexes/17575".
[2025-04-26T23:07:16Z INFO service::utils::clean] Find directory "pg_vectors/indexes/17575/segments/94627c6a-7461-4fac-aef8-bf0902748554".
2025-04-26 23:07:16.325 UTC [1] LOG: database system is ready to accept connections
2025-04-26 23:07:27.858 UTC [43] ERROR: relation "clip_index" does not exist
2025-04-26 23:07:27.858 UTC [43] STATEMENT: reindex index clip_index
2025-04-26 23:07:38.493 UTC [51] ERROR: relation "clip_index" does not exist
2025-04-26 23:07:38.493 UTC [51] STATEMENT: reindex index clip_index
2025-04-26 23:07:44.686 UTC [64] ERROR: relation "clip_index" does not exist
2025-04-26 23:07:44.686 UTC [64] STATEMENT: reindex index clip_index
2025-04-26 23:07:50.171 UTC [78] ERROR: relation "clip_index" does not exist
2025-04-26 23:07:50.171 UTC [78] STATEMENT: reindex index clip_index
2025-04-26 23:07:55.796 UTC [87] ERROR: relation "clip_index" does not exist
2025-04-26 23:07:55.796 UTC [87] STATEMENT: reindex index clip_index
Still don't know what happened, but I successfully restored yesterday's backup without issue
It seems the issue is you switched to a model that uses a different embedding vector length, and the database crapped itself
Did you re-run the machine learning on all assets after switching @disorderlyconduct ?
I did restart the Smart Search on all assets, and it seemed like it was working, but then I checked on it a few min later and it was done, far far too fast