# Ticket(s) Closed

- Closes #

## What

## Why

## How

## Tests

# Ticket(s) Closed

- Closes #

## What

## Why

## How

## Tests

)

## What

This removes the (sneaky) addition of an "average layer size" based on
the configured `target_segment_count` into the list of
`background_layer_sizes`.

The reason we're removing this is that when this happens on an existing,
large, unbalanced index, it will cause essentially the entire index to
be merged in the background, and that's almost never what one would want
in a production setting.

## Why

Looking to improve operational problems.

## How

## Tests

Co-authored-by: Eric Ridge <eebbrr@gmail.com>

## What

This is the 0.17.x-specific version of PR #3081.

## Why

Our tantivy timeline has forked and so has our dependency points.

## How

## Tests

…dexed fields… (#3089)

…). (#3045)

# Ticket(s) Closed

- Closes #3043 

## What

Prevents system crashes when `paradedb.enable_filter_pushdown` is
enabled and heap filter expressions contain subqueries.

## Why

When `enable_filter_pushdown` is enabled, expressions that cannot be
pushed down to the index are converted to `Qual::HeapExpr` for
heap-based evaluation. However, `HeapFieldFilter` uses
`CreateStandaloneExprContext()` which lacks the infrastructure needed
for subquery evaluation (subplans, proper executor state), causing
system crashes when subqueries are present.

## How

Added subquery detection in `try_pushdown()` using the existing
`contains_exec_param()` function to identify expressions containing EXEC
parameters (subqueries). When subqueries are detected, the function
returns `None` instead of creating a `Qual::HeapExpr`, causing
PostgreSQL's regular executor to handle the expression safely.

## Tests

- Added test cases in `unified_expression_comprehensive.sql` with
subqueries in heap filter expressions

---------

# Ticket(s) Closed

- Closes #

## What

## Why

## How

## Tests

Co-authored-by: Moe <mdashti@gmail.com>
Co-authored-by: Stu Hood <stuhood@paradedb.com>

## What

Teach the FSM how to conditionally upgrade its buffer lock to an
exclusive lock, skipping the page if it can't be acquired (unless it's
the last page, then we wait).

While here, centralize `std::thread::available_parallelism()` (it
actually showed up on profiles!) and gate some `pgrx::debug!()`
statements

## Why

Trying to remove some overhead that showed up during profiling.

## How

## Tests

Co-authored-by: Eric Ridge <eebbrr@gmail.com>

## What

Our recent changes to tantivy have adjusted the default MemoryArena page
size from 1MB to 512k, and this actually has a downstream effect of
allowing indexers to fit more docs in the arena before tantivy calcuates
that the indexer has exceeded its memory budget.

We have a test that exposed this, so this adjustes the number of
documents to insert to generate 5 segments.

## Why

## How

## Tests

Co-authored-by: Eric Ridge <eebbrr@gmail.com>

# Ticket(s) Closed

- Closes #

## What

## Why

## How

## Tests