You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Related issue: QA/QC communication repository's issue 115
Currently, when viewing rounds by cluster in the UI, the data is not displaying correctly.
This issue is caused by the following reason
To get the rounds by cluster, the DB is queried in the following order. (roundsByCluster GraphQL API)
Get model_id(i32) and batch_ts list(Vec<NaiveDateTime>) from the column_description table in postgres.
Get a Vec<ModelBatchInfo> from rocksdb's batch_info cf with model_id, batch_ts as keys.
One of the columns in the column_description table, batch_ts, is using the timestamp without time zone type and is stored with a precision down to micro second.
The keys used when inserting with batch_info cf are the model_id and the id (time of the event), and in the case of id, a timestamp with a nano second value.
That's why we can't find the value by looking up the batch_info cf with batch_ts as the key.
The text was updated successfully, but these errors were encountered:
kimhanbeom
changed the title
Data retrieval failure in batch_info due to mismatch between store key and search key types.
Data retrieval failure in batch_info due to mismatch between store key and search key types
Dec 16, 2024
kimhanbeom
changed the title
Data retrieval failure in batch_info due to mismatch between store key and search key types
Data retrieval failure in batch_info due to difference between the precision of the search key and the precision of the stored key
Dec 16, 2024
I believe this issue can be resolved by either adding a new column or changing the type of the existing column (batch_ts). If anyone has a better solution, please feel free to share it by commenting on this issue.
I appreciate your investigation and suggestions. Based on what you mentioned, I think it would be better to change the current type so as to support nano seconds.
@sehkone@syncpark@kimhanbeom
The batch_ts in the column_description table is of type timestamp without time zone, so it only stores values up to microseconds. The type of batch_ts can be changed to bigint to store values up to nanoseconds. This change requires the migration of existing data.
The migration process is expected to involve the following steps:
Related issue: QA/QC communication repository's issue 115
roundsByCluster
GraphQL API)model_id
(i32) andbatch_ts
list(Vec<NaiveDateTime>
) from thecolumn_description
table in postgres.Vec<ModelBatchInfo>
from rocksdb'sbatch_info
cf withmodel_id
,batch_ts
as keys.column_description
table,batch_ts
, is using thetimestamp without time zone
type and is stored with a precision down to micro second.batch_info
cf are themodel_id
and theid
(time of the event), and in the case of id, a timestamp with a nano second value.batch_info
cf withbatch_ts
as the key.The text was updated successfully, but these errors were encountered: