Hi Pinecone community,
I’m encountering two issues and would appreciate your help:
Pod Reduction Issue: I accidentally increased the number of pods through the dashboard, but now I find it impossible to reduce the pod count back to the original number. Is there any way to decrease the number of pods? What’s the recommended approach in this situation?
Pod Fullness Calculation: According to the documentation, an S1 pod with 1536-dimensional vectors should be able to store approximately 2.5M vectors. In my case:
Current record count: 266,880
Pod fullness: 70%
Vector dimensions: 1536
This raises several questions:
- Is this normal? The pod seems to be showing high fullness despite having relatively few vectors compared to the theoretical capacity
- What factors contribute to pod fullness calculation?
- What happens when pod fullness reaches 100%? Will I be unable to insert new vectors?
- Billing Questions: I have some additional questions about billing:
If I accidentally increase the number of pods, is this charged immediately in the billing cycle?
For new indexes: If I create a new index but delete it shortly after without using it, will I still be charged for this brief period?
How exactly does the billing cycle work in these scenarios?
I would greatly appreciate any clarification on these matters, particularly regarding immediate billing impact for pod changes and unused/quickly-deleted indexes.
Thank you in advance!
Hello @wangguoqing10110001,
Here are the steps to reduce the number of pods:
- Stop writing to the index
- Create a collection of the index
- Create a new index using the collection with the reduced number of pods
- Update the application to point to the new index.
- Enable writes
If you wish to use the same index name, you must delete the old one before you can reuse the same name.
Pod Fullness Calculation: According to the documentation, an S1 pod with 1536-dimensional vectors should be able to store approximately 2.5M vectors.
The example in the documentation assumes there is no metadata.
What factors contribute to pod fullness calculation?
This are the key factors:
- Vector ID length
- Dimensionality
- Size of metadata
- Number of Vectors
- Size of the sparse vector if being used
This is also an element of load that affects the fullness stat.
What happens when pod fullness reaches 100%? Will I be unable to insert new vectors?
All write operations will be blocked.
For more information about pod sizing please see - Choose a Pod Type and Size.
If I accidentally increase the number of pods, is this charged immediately in the billing cycle?
For pod-based indexes, the charge is per minute in 15-minute increments. When creating and quickly deleting the index, you will be charged for 15 minutes. For example, if a p1.x2 index were created and then deleted 10 minutes later, you would be changed for 30 minutes as the index has an effect pod size of two being a x2 size and then rounded up to the 15-minute mark.
15 minutes * 2 (number of affect pods) = 30 minute.
For more details, please refer to our documentation: Understanding Cost - Pod-based-Indexes