djg

Hi @darcythomas, > A policy that the wire protocol and the storage format are not changed in the same commit/release. > (Or have a backwards compatibility by version -1 of...

If `debug` is compiled in this flag should still be usable. But yeah, for the typical production build, we would probably elide `debug` at comptime.

Hi @joeky888! We don't plan to support multi-master. Our existing architecture is highly available though — if the primary fails, another replica will automatically take over. All writes, and all...

Yeah, TB is not horizontally scalable. But it will support very high + reliable throughput, so it probably won't be a bottleneck. For very expensive/complex queries, you would want to...

CDC isn't implemented yet, but we do have progress here: The underlying LSM query primitives are implemented (to support querying transfers for an account -- https://docs.tigerbeetle.com/reference/operations/get_account_transfers). (We are working to...

For object trees only: If we're willing to pay an extra indirection cost + complexity, the hash map could be `timestamp → usize` instead of `Value (implied timestamp key) →...

> store the values themselves in a fixed-capacity arraylist Which we could then sort in-place and swap with the immutable table's arraylist when we transition mutable → immutable to avoid...

W.r.t. hashing it looks like there's quite a bit of variance depending on the CPU: https://bench.cr.yp.to/results-hash.html It looks like Blake3 is a great choice on Intel, but not so good...

I'm not sure! It probably depends somewhat on the language in question, too.

In https://github.com/tigerbeetle/tigerbeetle/issues/1158#issuecomment-1690548154 Prior to the `superblock not found` error there was a `PathAlreadyExists` error -- make sure the data file doesn't already exist before you try to format it.