Nico/eng 2799 fix erlcass segfault3 #2

sneako · 2025-10-31T13:00:05Z

Note

Converts erlcass into an embeddable supervisor with exponential-backoff connection management and safer NIF prepare behavior, updates config to persistent_term, improves reconnection and ETS handling, and adjusts tests/benchmarks.

Runtime/OTP structure:
- Replace application module with supervisor erlcass_sup (embeddable via child_spec/1), add stop/0 and release_cluster/0.
- Introduce erlcass_connection_manager for cluster setup, monitoring, and exponential-backoff reconnect; prepares configured statements at startup.
- src/erlcass.erl: add reconnect scheduling (retry_ref, reconnect_delay), handle not-connected replies, and prepare cached statements on (re)connect.
NIF/C++ safety:
- Track session connection state in enif_cass_session; set on connect/close callbacks.
- Block cass_session_prepare when not connected, returning "no Cassandra connections available"; add extra allocation/error checks.
- Add constant kNoConnectionsAvailableMsg.
Configuration:
- New persistent_term-based config in erlcass_utils (set_config/1, get_config/0, clear_config/0), supports maps/proplists; default reconnect_delay (?RECONNECT_DELAY).
Caches/ETS:
- Make erlcass_stm_cache create idempotent; add erlcass_stm_sessions:reset/0 and idempotent create.
Tests/benchmarks:
- Common Test uses erlcass_sup:start_link/1 and stop/0; add no_connection_prepare_SUITE asserting prepare fails without connections.
- Benchmarks start via supervisor and pre-prepare statements when using erlcass.
Misc:
- erlcass.app.src: set {type, library}; add ?RECONNECT_DELAY in headers.
- Add AGENTS.md and extend .gitignore for CT artifacts.

^{Written by Cursor Bugbot for commit cd1afec. This will update automatically on new commits. Configure here.}

…tion available

cubic-dev-ai

4 issues found across 19 files

Prompt for AI agents (all 4 issues)


Understand the root cause of the following 4 issues and fix them.


<file name="c_src/nif_cass_session.cc">

<violation number="1" location="c_src/nif_cass_session.cc:90">
The async session callbacks now dereference cb-&gt;session-&gt;connected, but the callback only keeps a raw enif_cass_session* without enif_keep_resource. If Erlang GC drops the session term before the future completes, nif_cass_session_free can free that memory, so this write touches freed memory and can segfault.</violation>
</file>

<file name="src/erlcass.app.src">

<violation number="1" location="src/erlcass.app.src:11">
Replacing the application callback with `{type, library}` removes the `erlcass` application module. OTP will now treat `erlcass` as a pure library, so `application:start(erlcass)` fails with `{error,{bad_type,library}}` and the supervisor in `erlcass_sup` never starts. Please keep the `{mod, {erlcass_app, []}}` entry so the application can boot.</violation>
</file>

<file name="benchmarks/load_test.erl">

<violation number="1" location="benchmarks/load_test.erl:87">
Stopping and restarting the supervisor here clears the newly prepared `execute_query` statement, yet the code never re-prepares it before `async_execute/2` runs, so the benchmark will crash once the driver restarts.</violation>
</file>

<file name="src/erlcass.erl">

<violation number="1" location="src/erlcass.erl:522">
When `session_prepare_cached_statements/1` fails part-way, we close the new session but leave any already-inserted prepared statement entries in ETS pointing at that closed session. Please clear `erlcass_stm_sessions` (e.g., call `erlcass_stm_sessions:reset()` before returning) in both error branches so callers cannot bind/execute against stale session pointers.</violation>
</file>

_{React with 👍 or 👎 to teach cubic. Mention @cubic-dev-ai to give feedback, ask questions, or re-run the review.}

cubic-dev-ai · 2025-10-31T13:13:55Z

c_src/nif_cass_session.cc

+    {
        result = cass_future_error_to_nif_term(cb->env, future);
+        if(cb->session != NULL)
+            cb->session->connected = false;


The async session callbacks now dereference cb->session->connected, but the callback only keeps a raw enif_cass_session* without enif_keep_resource. If Erlang GC drops the session term before the future completes, nif_cass_session_free can free that memory, so this write touches freed memory and can segfault.

Prompt for AI agents

Address the following comment on c_src/nif_cass_session.cc at line 90: <comment>The async session callbacks now dereference cb->session->connected, but the callback only keeps a raw enif_cass_session* without enif_keep_resource. If Erlang GC drops the session term before the future completes, nif_cass_session_free can free that memory, so this write touches freed memory and can segfault.</comment> <file context> @@ -80,9 +84,17 @@ void on_session_connect(CassFuture* future, void* user_data) + { result = cass_future_error_to_nif_term(cb->env, future); + if(cb->session != NULL) + cb->session->connected = false; + } else </file context>

cubic-dev-ai · 2025-10-31T13:13:55Z

src/erlcass.app.src

        stdlib
    ]},
-    {mod, {erlcass_app, []}},
+    {type, library},


Replacing the application callback with {type, library} removes the erlcass application module. OTP will now treat erlcass as a pure library, so application:start(erlcass) fails with {error,{bad_type,library}} and the supervisor in erlcass_sup never starts. Please keep the {mod, {erlcass_app, []}} entry so the application can boot.

Prompt for AI agents

Address the following comment on src/erlcass.app.src at line 11: <comment>Replacing the application callback with `{type, library}` removes the `erlcass` application module. OTP will now treat `erlcass` as a pure library, so `application:start(erlcass)` fails with `{error,{bad_type,library}}` and the supervisor in `erlcass_sup` never starts. Please keep the `{mod, {erlcass_app, []}}` entry so the application can boot.</comment> <file context> @@ -8,7 +8,7 @@ stdlib ]}, - {mod, {erlcass_app, []}}, + {type, library}, {env, []}, </file context>

Suggested change

{type, library},

{mod, {erlcass_app, []}},

cubic-dev-ai · 2025-10-31T13:13:55Z

benchmarks/load_test.erl

    eprof:start_profiling([self()]),
-    ok = application:set_env(erlcass, keyspace, ?KEYSPACE),
-    start(),
+    ok = erlcass_sup:stop(),


Stopping and restarting the supervisor here clears the newly prepared execute_query statement, yet the code never re-prepares it before async_execute/2 runs, so the benchmark will crash once the driver restarts.

Prompt for AI agents

Address the following comment on benchmarks/load_test.erl at line 87: <comment>Stopping and restarting the supervisor here clears the newly prepared `execute_query` statement, yet the code never re-prepares it before `async_execute/2` runs, so the benchmark will crash once the driver restarts.</comment> <file context> @@ -81,8 +84,8 @@ do_profiling(NrProc, RequestsNr, RepeatNumber) -> eprof:start_profiling([self()]), - ok = application:set_env(erlcass, keyspace, ?KEYSPACE), - start(), + ok = erlcass_sup:stop(), + ok = start(#{keyspace => ?KEYSPACE}), </file context>

cubic-dev-ai · 2025-10-31T13:13:56Z

src/erlcass.erl

+                ok ->
+                    {ok, SessionRef};
+                {error, _} = Error ->
+                    _ = do_close(SessionRef, self(), ?CONNECT_TIMEOUT),


When session_prepare_cached_statements/1 fails part-way, we close the new session but leave any already-inserted prepared statement entries in ETS pointing at that closed session. Please clear erlcass_stm_sessions (e.g., call erlcass_stm_sessions:reset() before returning) in both error branches so callers cannot bind/execute against stale session pointers.

Prompt for AI agents

Address the following comment on src/erlcass.erl at line 522: <comment>When `session_prepare_cached_statements/1` fails part-way, we close the new session but leave any already-inserted prepared statement entries in ETS pointing at that closed session. Please clear `erlcass_stm_sessions` (e.g., call `erlcass_stm_sessions:reset()` before returning) in both error branches so callers cannot bind/execute against stale session pointers.</comment> <file context> @@ -479,6 +512,45 @@ session_prepare_cached_statements(SessionRef) -> + ok -> + {ok, SessionRef}; + {error, _} = Error -> + _ = do_close(SessionRef, self(), ?CONNECT_TIMEOUT), + Error; + Other -> </file context>

cubic-dev-ai

No issues found across 3 files

sneako · 2025-11-03T09:54:44Z

@codex review this PR

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

chatgpt-codex-connector · 2025-11-03T10:01:45Z

src/erlcass_connection_manager.erl

+schedule_retry(State0, Reason) ->
+    State1 = cancel_timer(clear_child(State0)),
+    ok = erlcass_sup:release_cluster(),
+    Delay = maps:get(backoff, State1),
+    Next = next_backoff(Delay, maps:get(max_backoff, State1)),


Avoid releasing cluster while erlcass session keeps retrying

When a connection attempt fails, schedule_retry/2 releases the global Cassandra cluster but leaves the existing erlcass gen_server running and reconnecting on its own timer. The reconnect path (handle_info({'timeout',…, reconnect}, State) → attempt_connect/0) uses cass_session_connect/2 which expects cassandra_data->cluster to be non‑NULL. After release_cluster/0 that pointer is freed and set to NULL, so the next retry from the still‑running session will call into the NIF with a freed cluster pointer, leading to a VM crash when Cassandra is down at startup. Either stop the erlcass child before calling release_cluster/0 or pause its reconnect timer until a new cluster is created.

Useful? React with 👍 / 👎.

sneako added 4 commits October 30, 2025 12:29

add AGENTS.md, update gitignore

b3cb5f0

add test to reproduce segfault when preparing statement but no connec…

2f8a1ac

…tion available

fix segfault when preparing query withput connection available

6967414

convert to supervisor

4e8560b

cubic-dev-ai bot reviewed Oct 31, 2025

View reviewed changes

allow statements to be prepared to be passed in to config

cd1afec

cubic-dev-ai bot reviewed Nov 3, 2025

View reviewed changes

chatgpt-codex-connector bot reviewed Nov 3, 2025

View reviewed changes

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Nico/eng 2799 fix erlcass segfault3 #2

Nico/eng 2799 fix erlcass segfault3 #2

Uh oh!

sneako commented Oct 31, 2025 •

edited by cursor bot

Loading

Uh oh!

cubic-dev-ai bot left a comment

Uh oh!

cubic-dev-ai bot Oct 31, 2025 •

edited

Loading

Uh oh!

cubic-dev-ai bot Oct 31, 2025 •

edited

Loading

Uh oh!

cubic-dev-ai bot Oct 31, 2025 •

edited

Loading

Uh oh!

cubic-dev-ai bot Oct 31, 2025 •

edited

Loading

Uh oh!

cubic-dev-ai bot left a comment

Uh oh!

sneako commented Nov 3, 2025

Uh oh!

chatgpt-codex-connector bot left a comment

Uh oh!

chatgpt-codex-connector bot Nov 3, 2025

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Nico/eng 2799 fix erlcass segfault3 #2

Are you sure you want to change the base?

Nico/eng 2799 fix erlcass segfault3 #2

Uh oh!

Conversation

sneako commented Oct 31, 2025 • edited by cursor bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

cubic-dev-ai bot left a comment

Choose a reason for hiding this comment

Uh oh!

cubic-dev-ai bot Oct 31, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

cubic-dev-ai bot Oct 31, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

cubic-dev-ai bot Oct 31, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

cubic-dev-ai bot Oct 31, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

cubic-dev-ai bot left a comment

Choose a reason for hiding this comment

Uh oh!

sneako commented Nov 3, 2025

Uh oh!

chatgpt-codex-connector bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

chatgpt-codex-connector bot Nov 3, 2025

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

sneako commented Oct 31, 2025 •

edited by cursor bot

Loading

cubic-dev-ai bot Oct 31, 2025 •

edited

Loading

cubic-dev-ai bot Oct 31, 2025 •

edited

Loading

cubic-dev-ai bot Oct 31, 2025 •

edited

Loading

cubic-dev-ai bot Oct 31, 2025 •

edited

Loading