Show Posts
All API services for testnet are breaking because of this. Chain.com, Blockcypher, Block.io. Blocktrail seems to be the only one to be keeping up but they dont have a proper API to make curl calls. The APIs show the following errors:
These issues obviously come from the Bitcoind instances dropping transactions and the databases that the APIs are using miss entries. I would say an API should be better than this, but whatever.
- No notification on new transaction
- Incorrect balance (up to negative balances) for addresses
- Denial of tx broadcast because "referenced transaction not known"
These issues obviously come from the Bitcoind instances dropping transactions and the databases that the APIs are using miss entries. I would say an API should be better than this, but whatever.
Hi levino, I'm sorry to hear you've been having issues with our (BlockCypher's) services on testnet. gmaxwell's suspicion is correct, we have experienced uncharacteristically long branches (100's of blocks) on testnet due in part to massive block throughput changes and spikes, which we are not as good at handling as we'd like (yet). The transient nature of many testnet peers exacerbates this problem, causing poor relay and a higher chance of miners working on branches and not the main chain.
On the one hand, this is giving us a great test bed for our own code base, and as a result we're steadily improving our performance in these corner cases, but as you've noticed they make our services fall behind for periods of time. Testnet is a protocol test bed, after all, and that means that all of us service providers are in the same boat as everyone else in experiencing new, bizarre, and extreme cases. Testnet is good for exercising issues in a realistic environment; it is not great for providing reliable, high availability services.
For trying out our services, we also provide our own testnet, which adheres to most of Bitcoin's parameters but better reflects its stability.
