Handle missing written_rows in insert

[hkrutzer]

This happens in the latest Clickhouse version (25.1)

It should resolve https://github.com/plausible/ecto_ch/actions/runs/13110662749/job/36573653799

[ruslandoga]

with falls through with the unmatched value (e.g. {:error, Jason.DecodeError.t()}). I think we might need to change the logic a bit to preserve num_rows :: integer | nil

[ruslandoga]

How about

num_rows =
  if summary = get_header(headers, "x-clickhouse-summary") do
    summary = Jason.decode!(summary)

    if written_rows = Map.get(summary, "written_rows") do
      String.to_integer(written_rows)
    end
  end

[hkrutzer]

Great catch. What do you think of:

    num_rows =
      with summary <- get_header(headers, "x-clickhouse-summary"),
           summary <- Jason.decode!(summary),
           written_rows when not is_nil(written_rows) <- Map.get(summary, "written_rows") do
        String.to_integer(written_rows)
      end

[ruslandoga]

I think it would "crash" on missing x-clickhouse-summary (get_header returns nil, with continues, Jason fails on nil) which might be OK. The initial goal was to allow missing x-clickhouse-summary but ensure it's JSON (raise otherwise). Not sure about how important it is now, though. Right now I think we probably shouldn't fail inserts at all on missing num_rows info and document it in ecto_ch that insert_all num_rows is unreliable.

[ruslandoga]

Thank you for the fix!

---

Strange, I can't reproduce the error locally on current master and ClickHouse local version 25.1.2.3 (official build). ClickHouse stopped returning written_rows in insert into ... select ... inserts.

---

And #237 passed as well. Ah, the failing CI is actually in ecto_ch! Sorry!

---

I have verified that the fix works, but I still think we need to be careful with with #236 (comment)

[ruslandoga]

Thank you! I'm going to cut a release right away!

---

Published as v0.3.0 since there have been some other changes since v0.2.x.

---

And I have also published v0.2.10 with just this change: https://diff.hex.pm/diff/ch/0.2.9..0.2.10

[hkrutzer]

Thanks, I did end up needing v0.2.10 because ecto_ch doesn't accept v0.3.0. That may not be intentional?

[ruslandoga]

Right, ecto_ch needs to be updated to allow ch 0.3.0: plausible/ecto_ch#221

---

UPDATE: I've released ecto_ch v0.6.0 which depends on ch v0.3.0.