https://github.com/apache/cassandra/pull/5063 Isn't that large? Especially compared to nearly 40k changes like https://github.com/apache/cassandra/pull/4967 that dont even have a JIRA attached
Long term goal is sub second splitting something that takes up to an hour of heavy CPU currently with zero disk IO seems valuable regardless. Making things like STCS, TWCS, and IR actually work (TWCS/STCS will not work with IR) Chris On Tue, Sep 15, 2026 at 5:23 AM Benedict Elliott Smith <[email protected]> wrote: > I think we need to settle a meta discussion first. We have had a > significant uptick in patches that are predominantly AI generated recently, > and especially for large patches like this, we need to agree a > project-level policy before any more are merged. I will send a separate > email in a few days with my thoughts to kick things off. I'm sure the > discussion will be… productive, but it is overdue. We all have enough > experience with the technology now, and no doubt have sufficiently > diverging viewpoints that for the project to function we must agree on a > shared understanding of what is expected and appropriate. > > Independent of this concern, regarding this patch I am in favour of "zero > copy" range streaming, but I am unclear whether the complexity of linking > file sub ranges is justified. > > > On 2026/09/14 19:22:43 Chris Lohfink wrote: > > Hi everyone, > > > > I'd like to open CEP-66, Zero-copy SSTable splitting, for discussion: > > > > * > https://cwiki.apache.org/confluence/spaces/CASSANDRA/pages/451972773/draft+CEP-66+Zero-copy+SSTable+splitting > > < > https://cwiki.apache.org/confluence/spaces/CASSANDRA/pages/451972773/draft+CEP-66+Zero-copy+SSTable+splitting > > > > * > > > > Anticompaction and partial-range streaming currently rewrite rows whose > > encoded representation already exists on disk. This consumes CPU, creates > > substantial heap churn and write amplification, and increases temporary > > disk pressure. > > > > CEP-66 proposes splitting eligible compressed SSTables by retaining > > contiguous runs of their existing compression chunks. Cassandra would > > rebuild the child SSTables' indexes and other derived components without > > deserializing, serializing, or recompressing their rows. > > > > "Zero-copy" here primarily means reusing the encoded bytes instead of > > rewriting rows. On filesystems that support range reflinks, Cassandra can > > also share the underlying extents meaning no new data written. Other > > filesystems, including ext4, would copy the already-compressed bytes and > > still avoid the row rewrite. > > > > The proposal is staged. It starts with an opt-in `sstablesplit > --zero-copy` > > mode for BIG-format SSTables in Cassandra 7.0/trunk. Later phases add BTI > > support, secondary indexes, anticompaction, and partial-range streaming. > > Existing implementations remain the default and provide the fallback for > > unsupported inputs. > > > > I'd particularly appreciate feedback on: > > > > - The retained-prefix representation and proposed Cassandra 7.0 SSTable > > format change > > - Rebuilding or conservatively deriving child metadata without decoding > rows > > - The integrity and performance tradeoff around `Digest.crc32` generation > > - The staged rollout, compatibility rules, and fallback behavior > > - Any correctness, operational, or filesystem concerns the proposal has > > missed > > > > Thanks, and I look forward to the discussion. > > > > Regards, > > Chris Lohfink > > >
