Is calibration net still supported? Providers are either failing to complete uploads entirely or failing the SP-to-SP replication pull for the secondary copy. I've had to drop to 1 copy and exclude specific provider IDs just to get uploads through.
By default, synapse-sdk stores 2 copies: one "primary" provider that the client uploads bytes to, plus a "secondary" that pulls from the primary via SP-to-SP transfer. The client upload only goes to the primary — if that primary is down or flaky, the whole upload throws StoreError: Failed to store on primary provider N and nothing commits. I've seen this specifically with provider 4 (caliberation-pdp.infrafolio.com) being selected as primary and failing at the store step's network request.
On calibration net, provider availability is inconsistent enough that the default 2-copy replication frequently fails at either the primary upload stage or the secondary SP-to-SP pull stage. This makes the network essentially unusable for development without manual workarounds.
The ask: Can calibration net providers be looked at? Either get the flaky ones back online or remove them from rotation so auto-selection stops picking dead providers as primary. Right now the default settings are broken enough that uploads just don't complete.
Is calibration net still supported? Providers are either failing to complete uploads entirely or failing the SP-to-SP replication pull for the secondary copy. I've had to drop to 1 copy and exclude specific provider IDs just to get uploads through.
By default, synapse-sdk stores 2 copies: one "primary" provider that the client uploads bytes to, plus a "secondary" that pulls from the primary via SP-to-SP transfer. The client upload only goes to the primary — if that primary is down or flaky, the whole upload throws StoreError: Failed to store on primary provider N and nothing commits. I've seen this specifically with provider 4 (caliberation-pdp.infrafolio.com) being selected as primary and failing at the store step's network request.
On calibration net, provider availability is inconsistent enough that the default 2-copy replication frequently fails at either the primary upload stage or the secondary SP-to-SP pull stage. This makes the network essentially unusable for development without manual workarounds.
The ask: Can calibration net providers be looked at? Either get the flaky ones back online or remove them from rotation so auto-selection stops picking dead providers as primary. Right now the default settings are broken enough that uploads just don't complete.