miri: Enable tests using epoll_wait

Based on recent segfaults we definitely want more Miri testing.

Author: def-

.config/nextest.toml

@@ -2,7 +2,7 @@
 slow-timeout = { period = "60s", terminate-after = 2 }
 
 [profile.default-miri]
-slow-timeout = { period = "600s", terminate-after = 2 }
+slow-timeout = { period = "1200s", terminate-after = 2 }
 
 # For a given configuration parameter, the first override to match wins. Keep
 # these sorted in order from most specific to least specific.

bin/ci-builder

@@ -180,6 +180,14 @@ case "$cmd" in
             --env NIGHTLY_CANARY_CONFLUENT_CLOUD_API_SECRET
             --env NO_COLOR
             --env PYPI_TOKEN
+            --env ZOOKEEPER_ADDR
+            --env KAFKA_ADDRS
+            --env SCHEMA_REGISTRY_URL
+            --env POSTGRES_URL
+            --env COCKROACH_URL
+            --env MZ_SOFT_ASSERTIONS
+            --env MZ_PERSIST_EXTERNAL_STORAGE_TEST_S3_BUCKET
+            --env MZ_PERSIST_EXTERNAL_STORAGE_TEST_POSTGRES_URL
         )
 
         if [[ $detach_container == "true" ]]; then

ci/test/cargo-test-miri.sh

@@ -18,5 +18,5 @@ set -euo pipefail
 # so keep them away from the `target` directory.
 export CARGO_TARGET_DIR="$PWD/miri-target"
 export MIRIFLAGS="-Zmiri-disable-isolation -Zmiri-strict-provenance"
-# exclude netwrok based tests, they mostly fail on epoll_wait
-cargo miri nextest run -j"$(nproc)" --no-fail-fast --workspace --exclude 'mz-adapter*' --exclude 'mz-environmentd*' --exclude 'mz-expr*' --exclude 'mz-compute-client*' --exclude 'mz-persist-client*' --exclude 'mz-ssh-util*' --exclude 'mz-rocksdb*'
+# exclude network-based and more complex tests which run out of memory
+cargo miri nextest run -j"$(nproc)" --no-fail-fast --workspace --exclude 'mz-environmentd*' --exclude 'mz-compute-client*' --exclude 'mz-ssh-util*' --exclude 'mz-rocksdb*'

ci/test/cargo-test/mzcompose.py

@@ -38,6 +38,7 @@
 
 
 def workflow_default(c: Composition, parser: WorkflowArgumentParser) -> None:
+    parser.add_argument("--miri", action="store_true")
     parser.add_argument("args", nargs="*")
     args = parser.parse_args()
     c.up("zookeeper", "kafka", "schema-registry", "postgres", "cockroach")
@@ -116,17 +117,23 @@ def workflow_default(c: Composition, parser: WorkflowArgumentParser) -> None:
             ],
             env=env,
         )
-        cpu_count = os.cpu_count()
-        assert cpu_count
-        spawn.runv(
-            [
-                "cargo",
-                "nextest",
-                "run",
-                "--profile=ci",
-                # Most tests don't use 100% of a CPU core, so run two tests per CPU.
-                f"--test-threads={cpu_count * 2}",
-                *args.args,
-            ],
-            env=env,
-        )
+        if args.miri:
+            spawn.runv(
+                ["bin/ci-builder", "run", "nightly", "ci/test/cargo-test-miri.sh"],
+                env=env,
+            )
+        else:
+            cpu_count = os.cpu_count()
+            assert cpu_count
+            spawn.runv(
+                [
+                    "cargo",
+                    "nextest",
+                    "run",
+                    "--profile=ci",
+                    # Most tests don't use 100% of a CPU core, so run two tests per CPU.
+                    f"--test-threads={cpu_count * 2}",
+                    *args.args,
+                ],
+                env=env,
+            )

ci/test/pipeline.template.yml

@@ -155,9 +155,24 @@ steps:
 
   - id: miri-test
     label: Miri test
-    command: bin/ci-builder run nightly ci/test/cargo-test-miri.sh
     inputs: [src]
-    timeout_in_minutes: 30
+    timeout_in_minutes: 60
+    artifact_paths: [junit_*.xml, target/nextest/ci/junit_cargo-test.xml]
+    inputs:
+      - Cargo.lock
+      - "**/Cargo.toml"
+      - "**/*.rs"
+      - "**/*.proto"
+      - "**/testdata/**"
+    env:
+      AWS_DEFAULT_REGION: "us-east-1"
+      # cargo-test's coverage is handled separately by cargo-llvm-cov
+      BUILDKITE_MZCOMPOSE_PLUGIN_SKIP_COVERAGE: "true"
+    plugins:
+      - ./ci/plugins/scratch-aws-access: ~
+      - ./ci/plugins/mzcompose:
+          composition: cargo-test
+          args: [--miri]
     agents:
       queue: builder-linux-x86_64
     coverage: skip

src/adapter/src/catalog.rs

@@ -8399,6 +8399,7 @@ mod tests {
     /// search paths, so do not require schema qualification on system objects such
     /// as types.
     #[mz_ore::test(tokio::test)]
+    #[cfg_attr(miri, ignore)] //  unsupported operation: can't call foreign function `TLS_client_method` on OS `linux`
     async fn test_minimal_qualification() {
         Catalog::with_debug(NOW_ZERO.clone(), |catalog| async move {
             struct TestCase {
@@ -8471,6 +8472,7 @@ mod tests {
     }
 
     #[mz_ore::test(tokio::test)]
+    #[cfg_attr(miri, ignore)] //  unsupported operation: can't call foreign function `TLS_client_method` on OS `linux`
     async fn test_catalog_revision() {
         let debug_stash_factory = DebugStashFactory::new().await;
         {
@@ -8506,6 +8508,7 @@ mod tests {
     }
 
     #[mz_ore::test(tokio::test)]
+    #[cfg_attr(miri, ignore)] // unsupported operation: can't call foreign function `TLS_client_method` on OS `linux`
     async fn test_effective_search_path() {
         Catalog::with_debug(NOW_ZERO.clone(), |catalog| async move {
             let mz_catalog_schema = (
@@ -8651,6 +8654,7 @@ mod tests {
     }
 
     #[mz_ore::test(tokio::test)]
+    #[cfg_attr(miri, ignore)] //  unsupported operation: can't call foreign function `TLS_client_method` on OS `linux`
     async fn test_builtin_migration() {
         enum ItemNamespace {
             System,
@@ -9305,6 +9309,7 @@ mod tests {
     }
 
     #[mz_ore::test(tokio::test)]
+    #[cfg_attr(miri, ignore)] //  unsupported operation: can't call foreign function `TLS_client_method` on OS `linux`
     async fn test_normalized_create() {
         Catalog::with_debug(NOW_ZERO.clone(), |catalog| {
             let catalog = catalog.for_system_session();
@@ -9489,6 +9494,7 @@ mod tests {
     }
 
     #[mz_ore::test(tokio::test)]
+    #[cfg_attr(miri, ignore)] //  unsupported operation: can't call foreign function `TLS_client_method` on OS `linux`
     async fn test_object_type() {
         let debug_stash_factory = DebugStashFactory::new().await;
         let stash = debug_stash_factory.open_debug().await;
@@ -9508,6 +9514,7 @@ mod tests {
     }
 
     #[mz_ore::test(tokio::test)]
+    #[cfg_attr(miri, ignore)] //  unsupported operation: can't call foreign function `TLS_client_method` on OS `linux`
     async fn test_get_privileges() {
         let debug_stash_factory = DebugStashFactory::new().await;
         let stash = debug_stash_factory.open_debug().await;

src/adapter/src/catalog/builtin.rs

@@ -4185,6 +4185,7 @@ mod tests {
     // Connect to a running Postgres server and verify that our builtin
     // types and functions match it, in addition to some other things.
     #[mz_ore::test(tokio::test)]
+    #[cfg_attr(miri, ignore)] //  unsupported operation: can't call foreign function `TLS_client_method` on OS `linux`
     async fn test_compare_builtins_postgres() {
         async fn inner(catalog: Catalog) {
             // Verify that all builtin functions:
@@ -4511,6 +4512,7 @@ mod tests {
 
     // Make sure pg views don't use types that only exist in Materialize.
     #[mz_ore::test(tokio::test)]
+    #[cfg_attr(miri, ignore)] //  unsupported operation: can't call foreign function `TLS_client_method` on OS `linux`
     async fn test_pg_views_forbidden_types() {
         Catalog::with_debug(SYSTEM_TIME.clone(), |catalog| async move {
             let conn_catalog = catalog.for_system_session();

src/adapter/src/config/params.rs

@@ -146,6 +146,7 @@ mod tests {
     use super::SynchronizedParameters;
 
     #[mz_ore::test]
+    #[cfg_attr(miri, ignore)] // unsupported operation: can't call foreign function `decNumberFromInt32` on OS `linux`
     fn test_github_18189() {
         let vars = SystemVars::default();
         let mut sync = SynchronizedParameters::new(vars);
@@ -156,6 +157,7 @@ mod tests {
     }
 
     #[mz_ore::test]
+    #[cfg_attr(miri, ignore)] // unsupported operation: can't call foreign function `decNumberFromInt32` on OS `linux`
     fn test_vars_are_synced() {
         let vars = SystemVars::default();
         let sync = SynchronizedParameters::new(vars);

src/adapter/src/coord/peek.rs

@@ -584,6 +584,7 @@ mod tests {
     use super::*;
 
     #[mz_ore::test]
+    #[cfg_attr(miri, ignore)] // unsupported operation: can't call foreign function `rust_psm_stack_pointer` on OS `linux`
     fn test_fast_path_plan_as_text() {
         let typ = RelationType::new(vec![ColumnType {
             scalar_type: ScalarType::String,

src/adapter/tests/parameters.rs

@@ -80,6 +80,7 @@ use mz_repr::ScalarType;
 use mz_sql::plan::PlanContext;
 
 #[mz_ore::test(tokio::test)]
+#[cfg_attr(miri, ignore)] // unsupported operation: can't call foreign function `TLS_client_method` on OS `linux`
 async fn test_parameter_type_inference() {
     let test_cases = vec![
         (

src/adapter/tests/sql.rs

@@ -102,6 +102,7 @@ use tokio::sync::Mutex;
 // catalog.
 
 #[mz_ore::test(tokio::test)]
+#[cfg_attr(miri, ignore)] // error: unsupported operation: can't call foreign function `TLS_client_method` on OS `linux`
 async fn datadriven() {
     datadriven::walk_async("tests/testdata/sql", |mut f| async {
         // The datadriven API takes an `FnMut` closure, and can't express to Rust that

src/adapter/tests/timestamp_selection.rs

@@ -261,6 +261,7 @@ fn parse_query_when(s: &str) -> QueryWhen {
 /// returns the chosen timestamp. Append `full` as an argument to it to see the entire
 /// TimestampDetermination.
 #[mz_ore::test]
+#[cfg_attr(miri, ignore)] // unsupported operation: can't call foreign function `decNumberFromInt32` on OS `linux`
 fn test_timestamp_selection() {
     datadriven::walk("tests/testdata/timestamp_selection", |tf| {
         let mut f = Frontiers {

src/compute-client/src/plan/join/mod.rs

@@ -421,6 +421,7 @@ mod tests {
         #![proptest_config(ProptestConfig::with_cases(32))]
 
         #[mz_ore::test]
+        #[cfg_attr(miri, ignore)] // error: unsupported operation: can't call foreign function `decContextDefault` on OS `linux`
         fn join_plan_protobuf_roundtrip(expect in any::<JoinPlan>() ) {
             let actual = protobuf_roundtrip::<_, ProtoJoinPlan>(&expect);
             assert!(actual.is_ok());

src/compute-client/src/plan/mod.rs

@@ -2197,6 +2197,7 @@ mod tests {
     proptest! {
         #![proptest_config(ProptestConfig::with_cases(10))]
         #[mz_ore::test]
+        #[cfg_attr(miri, ignore)] // error: unsupported operation: can't call foreign function `decContextDefault` on OS `linux`
         fn get_plan_protobuf_roundtrip(expect in any::<GetPlan>()) {
             let actual = protobuf_roundtrip::<_, ProtoGetPlan>(&expect);
             assert!(actual.is_ok());

src/compute-client/src/protocol/command.rs

@@ -512,6 +512,7 @@ mod tests {
         #![proptest_config(ProptestConfig::with_cases(32))]
 
         #[mz_ore::test]
+        #[cfg_attr(miri, ignore)] // error: unsupported operation: can't call foreign function `decContextDefault` on OS `linux`
         fn peek_protobuf_roundtrip(expect in any::<Peek>() ) {
             let actual = protobuf_roundtrip::<_, ProtoPeek>(&expect);
             assert!(actual.is_ok());

src/environmentd/tests/pgwire.rs

@@ -171,7 +171,6 @@ fn test_bind_params() {
 }
 
 #[mz_ore::test]
-#[cfg_attr(miri, ignore)] // unsupported operation: can't call foreign function `epoll_wait` on OS `linux`
 fn test_partial_read() {
     let server = util::start_server(util::Config::default()).unwrap();
     let mut client = server.connect(postgres::NoTls).unwrap();
@@ -198,7 +197,6 @@ fn test_partial_read() {
 }
 
 #[mz_ore::test]
-#[cfg_attr(miri, ignore)] // unsupported operation: can't call foreign function `epoll_wait` on OS `linux`
 fn test_read_many_rows() {
     let server = util::start_server(util::Config::default()).unwrap();
     let mut client = server.connect(postgres::NoTls).unwrap();
@@ -213,7 +211,6 @@ fn test_read_many_rows() {
 }
 
 #[mz_ore::test]
-#[cfg_attr(miri, ignore)] // unsupported operation: can't call foreign function `epoll_wait` on OS `linux`
 fn test_conn_startup() {
     let server = util::start_server(util::Config::default()).unwrap();
     let mut client = server.connect(postgres::NoTls).unwrap();
@@ -362,7 +359,6 @@ fn test_conn_startup() {
 }
 
 #[mz_ore::test]
-#[cfg_attr(miri, ignore)] // unsupported operation: can't call foreign function `epoll_wait` on OS `linux`
 fn test_conn_user() {
     let server = util::start_server(util::Config::default()).unwrap();
 
@@ -397,7 +393,6 @@ fn test_conn_user() {
 }
 
 #[mz_ore::test]
-#[cfg_attr(miri, ignore)] // unsupported operation: can't call foreign function `epoll_wait` on OS `linux`
 fn test_simple_query_no_hang() {
     let server = util::start_server(util::Config::default()).unwrap();
     let mut client = server.connect(postgres::NoTls).unwrap();
@@ -407,7 +402,6 @@ fn test_simple_query_no_hang() {
 }
 
 #[mz_ore::test]
-#[cfg_attr(miri, ignore)] // unsupported operation: can't call foreign function `epoll_wait` on OS `linux`
 fn test_copy() {
     let server = util::start_server(util::Config::default()).unwrap();
     let mut client = server.connect(postgres::NoTls).unwrap();
@@ -457,7 +451,6 @@ fn test_copy() {
 }
 
 #[mz_ore::test]
-#[cfg_attr(miri, ignore)] // unsupported operation: can't call foreign function `epoll_wait` on OS `linux`
 fn test_arrays() {
     let server = util::start_server(util::Config::default().unsafe_mode()).unwrap();
     let mut client = server.connect(postgres::NoTls).unwrap();
@@ -505,7 +498,6 @@ fn test_arrays() {
 }
 
 #[mz_ore::test]
-#[cfg_attr(miri, ignore)] // unsupported operation: can't call foreign function `epoll_wait` on OS `linux`
 fn test_record_types() {
     let server = util::start_server(util::Config::default()).unwrap();
     let mut client = server.connect(postgres::NoTls).unwrap();
@@ -573,15 +565,12 @@ fn pg_test_inner(dir: PathBuf, flags: &[&'static str]) {
 }
 
 #[mz_ore::test]
-#[cfg_attr(miri, ignore)] // unsupported operation: can't call foreign function `epoll_wait` on OS `linux`
 fn test_pgtest() {
     let dir: PathBuf = ["..", "..", "test", "pgtest"].iter().collect();
     pg_test_inner(dir, &[]);
 }
 
 #[mz_ore::test]
-// unsupported operation: can't call foreign function `epoll_wait` on OS `linux`
-#[cfg_attr(miri, ignore)]
 // Materialize's differences from Postgres' responses.
 fn test_pgtest_mz() {
     let dir: PathBuf = ["..", "..", "test", "pgtest-mz"].iter().collect();