Don't throw CURRENT_WRITE_BUFFER_IS_EXHAUSTED for normal behaviour

[Algunenano]

Changelog category (leave one):

• Improvement

Changelog entry (a user-readable short description of the changes that goes to CHANGELOG.md):

• Don't throw CURRENT_WRITE_BUFFER_IS_EXHAUSTED for normal behaviour

Documentation entry for user-facing changes

Information about CI checks: https://clickhouse.com/docs/en/development/continuous-integration/

While testing 23.3 settings http_wait_end_of_query and http_response_buffer_size I noticed the following error on system.errors only (queries finished correcly):

Row 1:
──────
server_shard:       1
version:            23.3.1.2778
name:               CURRENT_WRITE_BUFFER_IS_EXHAUSTED
code:               362
quantity:           20
last_error_message: MemoryWriteBuffer limit is exhausted
traceback:          DB::Exception::Exception(DB::Exception::MessageMasked&&, int, bool)
DB::MemoryWriteBuffer::addChunk()
DB::CascadeWriteBuffer::nextImpl()
DB::ParallelFormattingOutputFormat::collectorThreadFunction(std::__1::shared_ptr<DB::ThreadGroupStatus> const&)
void std::__1::__function::__policy_invoker<void ()>::__call_impl<std::__1::__function::__default_alloc_func<ThreadFromGlobalPoolImpl<true>::ThreadFromGlobalPoolImpl<DB::ParallelFormattingOutputFormat::ParallelFormattingOutputFormat(DB::ParallelFormattingOutputFormat::Params)::'lambda'()>(DB::ParallelFormattingOutputFormat::ParallelFormattingOutputFormat(DB::ParallelFormattingOutputFormat::Params)::'lambda'()&&)::'lambda'(), void ()>>(std::__1::__function::__policy_storage const*)
ThreadPoolImpl<std::__1::thread>::worker(std::__1::__list_iterator<std::__1::thread, void*>)

Looking at the exception it seems that CH uses CURRENT_WRITE_BUFFER_IS_EXHAUSTED to internally flag in CascadeWriteBuffer that it needs a new buffer. The problem is that using DB::Exception also reports this error to system.errors which is confusing since this is not an actual error.

Instead, use a custom exception to handle it this and only throw CURRENT_WRITE_BUFFER_IS_EXHAUSTED when there are no buffers anymore.

[alexey-milovidov]

Ok, but can we avoid using exceptions in this case?

[Algunenano]

AFAICS, only by rewriting CascadeWriteBuffer completely and changing WriteBuffer::next(), which is used pretty much everywhere, so it can flag this situation.