miri: Try enabling epoll_wait

Author: def-

.config/nextest.toml

@@ -12,15 +12,29 @@ filter = "package(mz-environmentd) and test(test_pgtest)"
 threads-required = 8
 slow-timeout = { period = "300s", terminate-after = 2 }
 
+[[profile.default-miri.overrides]]
+filter = "package(mz-environmentd) and test(test_pgtest)"
+threads-required = 8
+slow-timeout = { period = "3000s", terminate-after = 2 }
+
 [[profile.default.overrides]]
 filter = "package(mz-environmentd)"
 threads-required = 8
 slow-timeout = { period = "120s", terminate-after = 2 }
 
+[[profile.default-miri.overrides]]
+filter = "package(mz-environmentd)"
+threads-required = 8
+slow-timeout = { period = "1200s", terminate-after = 2 }
+
 [[profile.default.overrides]]
 filter = "package(mz-stash) and test(proptest_stash_migrate_json_to_proto)"
 slow-timeout = { period = "120s", terminate-after = 2 }
 
+[[profile.default-miri.overrides]]
+filter = "package(mz-stash) and test(proptest_stash_migrate_json_to_proto)"
+slow-timeout = { period = "1200s", terminate-after = 2 }
+
 [profile.ci]
 junit = { path = "junit_cargo-test.xml" }
 fail-fast = false

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 +18,4 @@ 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*'
+cargo miri nextest run -j"$(nproc)" --no-fail-fast --workspace

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 +117,20 @@ 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/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/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();

src/environmentd/tests/server.rs

@@ -243,7 +243,6 @@ fn test_source_sink_size_required() {
 
 // Test the POST and WS server endpoints.
 #[mz_ore::test]
-#[cfg_attr(miri, ignore)] // unsupported operation: can't call foreign function `epoll_wait` on OS `linux`
 fn test_http_sql() {
     // Datadriven directives for WebSocket are "ws-text" and "ws-binary" to send
     // text or binary websocket messages that are the input. Output is
@@ -347,7 +346,6 @@ fn test_http_sql() {
 
 // Test that the server properly handles cancellation requests.
 #[mz_ore::test]
-#[cfg_attr(miri, ignore)] // unsupported operation: can't call foreign function `epoll_wait` on OS `linux`
 fn test_cancel_long_running_query() {
     let config = util::Config::default().unsafe_mode();
     let server = util::start_server(config).unwrap();
@@ -459,19 +457,16 @@ fn test_cancellation_cancels_dataflows(query: &str) {
 
 // Test that dataflow uninstalls cancelled peeks.
 #[mz_ore::test]
-#[cfg_attr(miri, ignore)] // unsupported operation: can't call foreign function `epoll_wait` on OS `linux`
 fn test_cancel_dataflow_removal() {
     test_cancellation_cancels_dataflows("SELECT * FROM t AS OF 9223372036854775807");
 }
 
 #[mz_ore::test]
-#[cfg_attr(miri, ignore)] // unsupported operation: can't call foreign function `epoll_wait` on OS `linux`
 fn test_cancel_long_select() {
     test_cancellation_cancels_dataflows("WITH MUTUALLY RECURSIVE flip(x INTEGER) AS (VALUES(1) EXCEPT ALL SELECT * FROM flip) SELECT * FROM flip;");
 }
 
 #[mz_ore::test]
-#[cfg_attr(miri, ignore)] // unsupported operation: can't call foreign function `epoll_wait` on OS `linux`
 fn test_cancel_insert_select() {
     test_cancellation_cancels_dataflows("INSERT INTO t WITH MUTUALLY RECURSIVE flip(x INTEGER) AS (VALUES(1) EXCEPT ALL SELECT * FROM flip) SELECT * FROM flip;");
 }
@@ -566,13 +561,11 @@ fn test_closing_connection_cancels_dataflows(query: String) {
 }
 
 #[mz_ore::test]
-#[cfg_attr(miri, ignore)] // unsupported operation: can't call foreign function `epoll_wait` on OS `linux`
 fn test_closing_connection_for_long_select() {
     test_closing_connection_cancels_dataflows("WITH MUTUALLY RECURSIVE flip(x INTEGER) AS (VALUES(1) EXCEPT ALL SELECT * FROM flip) SELECT * FROM flip;".to_string())
 }
 
 #[mz_ore::test]
-#[cfg_attr(miri, ignore)] // unsupported operation: can't call foreign function `epoll_wait` on OS `linux`
 fn test_closing_connection_for_insert_select() {
     test_closing_connection_cancels_dataflows("CREATE TABLE t1 (a int); INSERT INTO t1 WITH MUTUALLY RECURSIVE flip(x INTEGER) AS (VALUES(1) EXCEPT ALL SELECT * FROM flip) SELECT * FROM flip;".to_string())
 }
@@ -920,7 +913,6 @@ fn test_old_storage_usage_records_are_reaped_on_restart() {
 }
 
 #[mz_ore::test]
-#[cfg_attr(miri, ignore)] // unsupported operation: can't call foreign function `epoll_wait` on OS `linux`
 fn test_default_cluster_sizes() {
     let config = util::Config::default()
         .with_builtin_cluster_replica_size("1".to_string())
@@ -952,7 +944,6 @@ fn test_default_cluster_sizes() {
 }
 
 #[mz_ore::test]
-#[cfg_attr(miri, ignore)] // unsupported operation: can't call foreign function `epoll_wait` on OS `linux`
 fn test_max_request_size() {
     let statement = "SELECT $1::text";
     let statement_size = statement.bytes().count();
@@ -1088,7 +1079,6 @@ fn test_max_statement_batch_size() {
 }
 
 #[mz_ore::test]
-#[cfg_attr(miri, ignore)] // unsupported operation: can't call foreign function `epoll_wait` on OS `linux`
 fn test_mz_system_user_admin() {
     let config = util::Config::default();
     let server = util::start_server(config).unwrap();

src/expr/src/interpret.rs

@@ -1212,6 +1212,7 @@ mod tests {
     }
 
     #[mz_ore::test]
+    #[cfg_attr(miri, ignore)] // error: unsupported operation: can't call foreign function `decContextDefault` on OS `linux`
     fn test_equivalence() {
         fn check(data: ExpressionData) -> Result<(), TestCaseError> {
             let ExpressionData {
@@ -1385,6 +1386,7 @@ mod tests {
     }
 
     #[mz_ore::test]
+    #[cfg_attr(miri, ignore)] // error: unsupported operation: can't call foreign function `decNumberFromInt32` on OS `linux`
     fn test_jsonb() {
         let arena = RowArena::new();
 

src/expr/src/scalar/mod.rs

@@ -2791,6 +2791,7 @@ mod tests {
     use super::*;
 
     #[mz_ore::test]
+    #[cfg_attr(miri, ignore)] // error: unsupported operation: can't call foreign function `rust_psm_stack_pointer` on OS `linux`
     fn test_reduce() {
         let relation_type = vec![
             ScalarType::Int64.nullable(true),

src/persist-client/src/batch.rs

@@ -869,7 +869,6 @@ mod tests {
     use super::*;
 
     #[mz_ore::test(tokio::test)]
-    #[cfg_attr(miri, ignore)] // unsupported operation: can't call foreign function `epoll_wait` on OS `linux`
     async fn batch_builder_flushing() {
         let data = vec![
             (("1".to_owned(), "one".to_owned()), 1, 1),
@@ -962,7 +961,6 @@ mod tests {
     }
 
     #[mz_ore::test(tokio::test)]
-    #[cfg_attr(miri, ignore)] // unsupported operation: can't call foreign function `epoll_wait` on OS `linux`
     async fn batch_builder_keys() {
         let cache = PersistClientCache::new_no_metrics();
         // Set blob_target_size to 0 so that each row gets forced into its own batch part
@@ -1004,7 +1002,6 @@ mod tests {
     }
 
     #[mz_ore::test(tokio::test)]
-    #[cfg_attr(miri, ignore)] // unsupported operation: can't call foreign function `epoll_wait` on OS `linux`
     async fn batch_builder_partial_order() {
         let cache = PersistClientCache::new_no_metrics();
         // Set blob_target_size to 0 so that each row gets forced into its own batch part

src/persist-client/src/cache.rs

@@ -656,7 +656,6 @@ mod tests {
     use super::*;
 
     #[mz_ore::test(tokio::test)]
-    #[cfg_attr(miri, ignore)] // unsupported operation: can't call foreign function `epoll_wait` on OS `linux`
     async fn client_cache() {
         let cache = PersistClientCache::new(
             PersistConfig::new(&DUMMY_BUILD_INFO, SYSTEM_TIME.clone()),

src/persist-client/src/critical.rs

@@ -377,7 +377,6 @@ mod tests {
     }
 
     #[mz_ore::test(tokio::test)]
-    #[cfg_attr(miri, ignore)] // unsupported operation: can't call foreign function `epoll_wait` on OS `linux`
     async fn rate_limit() {
         let client = crate::tests::new_test_client().await;
 
@@ -405,7 +404,6 @@ mod tests {
 
     // Verifies that the handle updates its view of the opaque token correctly
     #[mz_ore::test(tokio::test)]
-    #[cfg_attr(miri, ignore)] // unsupported operation: can't call foreign function `epoll_wait` on OS `linux`
     async fn handle_opaque_token() {
         let client = new_test_client().await;
         let shard_id = ShardId::new();

src/persist-client/src/internal/compact.rs

@@ -996,7 +996,6 @@ mod tests {
     // made it to main) where batches written by compaction would always have a
     // since of the minimum timestamp.
     #[mz_ore::test(tokio::test)]
-    #[cfg_attr(miri, ignore)] // unsupported operation: can't call foreign function `epoll_wait` on OS `linux`
     async fn regression_minimum_since() {
         let data = vec![
             (("0".to_owned(), "zero".to_owned()), 0, 1),
@@ -1064,7 +1063,6 @@ mod tests {
     }
 
     #[mz_ore::test(tokio::test)]
-    #[cfg_attr(miri, ignore)] // unsupported operation: can't call foreign function `epoll_wait` on OS `linux`
     async fn compaction_partial_order() {
         let data = vec![
             (
@@ -1150,7 +1148,6 @@ mod tests {
     }
 
     #[mz_ore::test(tokio::test)]
-    #[cfg_attr(miri, ignore)] // unsupported operation: can't call foreign function `epoll_wait` on OS `linux`
     async fn prefetches() {
         let desc = Description::new(
             Antichain::from_elem(0u64),