adriangb · adriangb · Jul 31, 2026 · Jul 31, 2026
diff --git a/README.md b/README.md
@@ -129,6 +129,33 @@ changed:
 
 Per-side env vars override shared env vars when both set the same key.
 
+### What ran: the Run configuration block
+
+Every comment the runner posts — running, completed, and failed — states what it
+is benchmarking: a `Comparing <changed> (<sha>) to <baseline> [diff]` line, and a
+collapsed **Run configuration** block holding the refs and env vars in the same
+YAML the trigger accepts.
+
+It is re-rendered from the job rather than quoted from the trigger comment,
+because the two differ: one `run benchmark tpch clickbench_1` fans out into one
+job (and one comment) per name, and a bare `run benchmarks` expands to the repo's
+default suite. The block shows what *this* run did, so pasting it back into a PR
+comment reproduces that run on its own.
+
+```yaml
+run benchmark tpch
+env:
+  CARGO_BUILD_JOBS: "1"
+baseline:
+  ref: "v45.0.0"
+changed:
+  env:
+    DATAFUSION_RUNTIME_MEMORY_LIMIT: "2G"
+```
+
+Sections with nothing to report are omitted, so an unconfigured run renders as
+just its trigger line.
+
 ### Peak memory pool reservation
 
 DataFusion records the high-water mark of `MemoryPool::reserved()` per query and

diff --git a/controller/src/bin/runner.rs b/controller/src/bin/runner.rs
@@ -9,7 +9,7 @@ use tracing::{error, info};
 use benchmark_controller::github;
 use benchmark_controller::runner::config::{BenchType, RunnerConfig};
 use benchmark_controller::runner::poster::CommentPoster;
-use benchmark_controller::runner::{bench_arrow, bench_datafusion, shell};
+use benchmark_controller::runner::{bench_arrow, bench_datafusion, shell, trigger};
 
 #[tokio::main]
 async fn main() {
@@ -68,9 +68,18 @@ async fn run_benchmark(config: &RunnerConfig, poster: &CommentPoster) -> Result<
 async fn post_error_comment(config: &RunnerConfig, poster: &CommentPoster) {
     let tail = shell::tail_log(20).await;
 
+    // A failure can happen before either checkout resolves, so there is no
+    // comparison line here — just the requested configuration.
+    let bench_names = match config.bench_type {
+        BenchType::ArrowCriterion => config.bench_name.as_str(),
+        _ => config.benchmarks.as_str(),
+    };
+    let config_block = trigger::config_block(config, bench_names);
+
     let footer = github::issues_footer(config.runner_repo_url.as_deref());
     let body = format!(
         "Benchmark for [this request]({}) failed.\n\n\
+         {config_block}\
          Last 20 lines of output:\n\
          <details><summary>Click to expand</summary>\n\n\
          ```\n\

diff --git a/controller/src/job_manager.rs b/controller/src/job_manager.rs
@@ -414,6 +414,14 @@ async fn create_k8s_job(
         env.push(env_var(k, v));
     }
 
+    // Also pass the shared block as a JSON map. The pod already has these set
+    // individually (above); the map lets the runner report them in PR comments
+    // without having to guess which of its env vars came from the trigger.
+    env.push(env_var(
+        "SHARED_ENV_VARS",
+        serde_json::to_string(&shared_env_vars)?,
+    ));
+
     // Pass per-side env vars as JSON maps for the runner to apply
     env.push(env_var("BASELINE_ENV_VARS", &job.baseline_env_vars));
     env.push(env_var("CHANGED_ENV_VARS", &job.changed_env_vars));

diff --git a/controller/src/runner/bench_arrow.rs b/controller/src/runner/bench_arrow.rs
@@ -11,6 +11,7 @@ use crate::runner::git;
 use crate::runner::monitor;
 use crate::runner::poster::CommentPoster;
 use crate::runner::shell;
+use crate::runner::trigger;
 
 /// Run an arrow-rs criterion benchmark comparing a PR branch to its merge-base.
 pub async fn run(config: &RunnerConfig, poster: &CommentPoster) -> Result<()> {
@@ -75,6 +76,15 @@ pub async fn run(config: &RunnerConfig, poster: &CommentPoster) -> Result<()> {
     } else {
         format!("{} (merge-base)", &base_sha[..7.min(base_sha.len())])
     };
+    let comparison = trigger::Comparison {
+        repo: &config.repo,
+        changed_display,
+        changed_sha: &changed_sha,
+        baseline_label: &baseline_label,
+        base_sha: &base_sha,
+    };
+    let config_block = trigger::config_block(config, bench_name);
+
     let footer = github::issues_footer(config.runner_repo_url.as_deref());
     let running_body = format!(
         "\u{1f916} Arrow criterion benchmark running (GKE) | [trigger]({})\n\
@@ -84,14 +94,12 @@ pub async fn run(config: &RunnerConfig, poster: &CommentPoster) -> Result<()> {
          {lscpu}\n\
          ```\n\n\
          </details>\n\n\
-         Comparing {changed_display} ({changed_sha}) to {baseline_label} \
-         [diff](https://github.com/{repo}/compare/{base_sha}..{changed_sha})\n\
-         BENCH_NAME={bench_name}\n\
+         {comparison}\n\n\
+         {config_block}\
          BENCH_COMMAND={bench_command_display}\n\
-         BENCH_FILTER={bench_filter}\n\
          Results will be posted here when complete{footer}",
         config.comment_url,
-        repo = config.repo,
+        comparison = comparison.line(),
     );
     let pr_number = config.pr_number()?;
     poster
@@ -193,6 +201,8 @@ pub async fn run(config: &RunnerConfig, poster: &CommentPoster) -> Result<()> {
         );
         format_result_comment(
             &config.comment_url,
+            &comparison.line(),
+            &config_block,
             &report,
             &resource_section,
             &instance_type,
@@ -209,6 +219,8 @@ pub async fn run(config: &RunnerConfig, poster: &CommentPoster) -> Result<()> {
             monitor::format_resource_comment("branch", &branch_stats).to_string();
         format_branch_only_result_comment(
             &config.comment_url,
+            &comparison.line(),
+            &config_block,
             &report,
             &resource_section,
             &instance_type,
@@ -253,8 +265,14 @@ async fn copy_criterion_baselines(base_dir: &Path, branch_dir: &Path) {
 }
 
 /// Format the result comment body.
+///
+/// `comparison` and `config_block` restate what was run, so the result reads on
+/// its own rather than only linking back to the trigger comment.
+#[allow(clippy::too_many_arguments)]
 fn format_result_comment(
     comment_url: &str,
+    comparison: &str,
+    config_block: &str,
     report: &str,
     resource_section: &str,
     instance_type: &str,
@@ -265,6 +283,8 @@ fn format_result_comment(
     format!(
         "\u{1f916} Arrow criterion benchmark completed (GKE) | [trigger]({comment_url})\n\n\
          **Instance:** `{instance_type}` ({pod_resources})\n\n\
+         {comparison}\n\n\
+         {config_block}\
          <details><summary>CPU Details (lscpu)</summary>\n\n\
          ```\n\
          {lscpu}\n\
@@ -285,8 +305,11 @@ fn format_result_comment(
 }
 
 /// Format the result comment body for branch-only runs (no baseline comparison).
+#[allow(clippy::too_many_arguments)]
 fn format_branch_only_result_comment(
     comment_url: &str,
+    comparison: &str,
+    config_block: &str,
     report: &str,
     resource_section: &str,
     instance_type: &str,
@@ -297,6 +320,8 @@ fn format_branch_only_result_comment(
     format!(
         "\u{1f916} Arrow criterion benchmark completed (GKE) | [trigger]({comment_url})\n\n\
          **Instance:** `{instance_type}` ({pod_resources})\n\n\
+         {comparison}\n\n\
+         {config_block}\
          <details><summary>CPU Details (lscpu)</summary>\n\n\
          ```\n\
          {lscpu}\n\
@@ -340,10 +365,17 @@ mod tests {
         );
     }
 
+    /// Stand-ins for the two "what was run" sections the runner passes in.
+    const COMPARISON: &str =
+        "Comparing my-branch (aaa) to bbb (merge-base) [diff](https://example.com/diff)";
+    const CONFIG_BLOCK: &str = "<details><summary>Run configuration</summary>\n\n```yaml\nrun benchmark concatenate_kernel\n```\n\n</details>\n\n";
+
     #[test]
     fn result_comment_format() {
         let comment = format_result_comment(
             "https://example.com/comment",
+            COMPARISON,
+            CONFIG_BLOCK,
             "test report\n",
             "resources\n",
             "c4a-standard-48",
@@ -358,12 +390,16 @@ mod tests {
         assert!(comment.contains("c4a-standard-48"));
         assert!(comment.contains("12 vCPU / 65 GiB"));
         assert!(comment.contains("lscpu output"));
+        assert!(comment.contains("Comparing my-branch (aaa)"));
+        assert!(comment.contains("run benchmark concatenate_kernel"));
     }
 
     #[test]
     fn branch_only_result_comment_format() {
         let comment = format_branch_only_result_comment(
             "https://example.com/comment",
+            COMPARISON,
+            CONFIG_BLOCK,
             "branch report\n",
             "branch resources\n",
             "c4a-standard-48",
@@ -372,6 +408,8 @@ mod tests {
             "",
         );
         assert!(comment.contains("Arrow criterion benchmark completed"));
+        assert!(comment.contains("Comparing my-branch (aaa)"));
+        assert!(comment.contains("run benchmark concatenate_kernel"));
         assert!(comment.contains("New benchmark — branch-only results"));
         assert!(comment.contains("[trigger](https://example.com/comment)"));
         assert!(comment.contains("branch report"));

diff --git a/controller/src/runner/bench_datafusion.rs b/controller/src/runner/bench_datafusion.rs
@@ -33,6 +33,7 @@ use crate::runner::monitor::{self, ResourceStats};
 use crate::runner::pool_peak::{self, BenchPeaks};
 use crate::runner::poster::CommentPoster;
 use crate::runner::shell;
+use crate::runner::trigger;
 
 /// Run DataFusion benchmarks comparing a PR branch to its merge-base.
 pub async fn run(config: &RunnerConfig, poster: &CommentPoster) -> Result<()> {
@@ -121,6 +122,15 @@ pub async fn run(config: &RunnerConfig, poster: &CommentPoster) -> Result<()> {
         format!("{} (merge-base)", &base_sha[..7.min(base_sha.len())])
     };
 
+    let comparison = trigger::Comparison {
+        repo: &config.repo,
+        changed_display,
+        changed_sha: &changed_sha,
+        baseline_label: &baseline_label,
+        base_sha: &base_sha,
+    };
+    let config_block = trigger::config_block(config, benchmarks);
+
     let footer = github::issues_footer(config.runner_repo_url.as_deref());
     let running_body = format!(
         "\u{1f916} Benchmark running (GKE) | [trigger]({})\n\
@@ -130,12 +140,11 @@ pub async fn run(config: &RunnerConfig, poster: &CommentPoster) -> Result<()> {
          {lscpu}\n\
          ```\n\n\
          </details>\n\n\
-         Comparing {changed_display} ({changed_sha}) to {baseline_label} \
-         [diff](https://github.com/{repo}/compare/{base_sha}..{changed_sha}) \
-         using: {benchmarks}\n\
+         {comparison}\n\n\
+         {config_block}\
          Results will be posted here when complete{footer}",
         config.comment_url,
-        repo = config.repo,
+        comparison = comparison.line(),
     );
     poster
         .post_comment(&config.repo, pr_number, &running_body)
@@ -339,6 +348,8 @@ pub async fn run(config: &RunnerConfig, poster: &CommentPoster) -> Result<()> {
     );
     let result_body = format_result_comment(
         &config.comment_url,
+        &comparison.line(),
+        &config_block,
         &report,
         &resource_section,
         &pool_section,
@@ -722,13 +733,19 @@ fn format_resource_section(
 
 /// Format the result comment body.
 ///
+/// `comparison` and `config_block` restate what was run — the same two lines
+/// the "running" comment opened with — so the result stands on its own instead
+/// of only linking back to the trigger.
+///
 /// `pool_section` is empty whenever no side recorded a `pool_peak_bytes` — the
 /// default, since runs set no memory limit unless the trigger comment asks for
 /// one. Its `<details>` block is then omitted entirely, leaving the comment as
 /// it was before the section existed.
 #[allow(clippy::too_many_arguments)]
 fn format_result_comment(
     comment_url: &str,
+    comparison: &str,
+    config_block: &str,
     report: &str,
     resource_section: &str,
     pool_section: &str,
@@ -749,6 +766,8 @@ fn format_result_comment(
     format!(
         "\u{1f916} Benchmark completed (GKE) | [trigger]({comment_url})\n\n\
          **Instance:** `{instance_type}` ({pod_resources})\n\n\
+         {comparison}\n\n\
+         {config_block}\
          <details><summary>CPU Details (lscpu)</summary>\n\n\
          ```\n\
          {lscpu}\n\
@@ -777,6 +796,8 @@ mod tests {
     fn result_comment_format() {
         let comment = format_result_comment(
             "https://example.com/comment",
+            "Comparing my-branch (aaa) to bbb (merge-base) [diff](https://example.com/diff)",
+            "<details><summary>Run configuration</summary>\n\n```yaml\nrun benchmark tpch\n```\n\n</details>\n\n",
             "test report\n",
             "resources\n",
             "",
@@ -798,10 +819,35 @@ mod tests {
         assert!(!comment.contains("Memory Pool Peaks"));
     }
 
+    #[test]
+    fn result_comment_restates_what_was_run() {
+        // A result comment should be readable without opening the trigger.
+        let comment = format_result_comment(
+            "https://example.com/comment",
+            "Comparing my-branch (aaa) to bbb (merge-base) [diff](https://example.com/diff)",
+            "<details><summary>Run configuration</summary>\n\n```yaml\nrun benchmark tpch\nbaseline:\n  ref: \"v45.0.0\"\n```\n\n</details>\n\n",
+            "test report\n",
+            "resources\n",
+            "",
+            "c4a-standard-48",
+            "12 vCPU / 65 GiB",
+            "lscpu output",
+            "",
+        );
+        assert!(comment.contains("Comparing my-branch (aaa) to bbb (merge-base)"));
+        assert!(comment.contains("<details><summary>Run configuration</summary>"));
+        assert!(comment.contains("run benchmark tpch"));
+        assert!(comment.contains("ref: \"v45.0.0\""));
+        // Stated up front, before the results themselves.
+        assert!(comment.find("Run configuration").unwrap() < comment.find("test report").unwrap());
+    }
+
     #[test]
     fn result_comment_includes_pool_section_when_recorded() {
         let comment = format_result_comment(
             "https://example.com/comment",
+            "Comparing my-branch (aaa) to bbb (merge-base) [diff](https://example.com/diff)",
+            "",
             "test report\n",
             "resources\n",
             "pool peaks table\n",

diff --git a/controller/src/runner/config.rs b/controller/src/runner/config.rs
@@ -3,7 +3,8 @@
 //! The controller passes these env vars to PR-triggered runner pods (see
 //! `job_manager.rs`): `PR_URL`, `COMMENT_ID`, `BENCHMARKS`, `BENCH_TYPE`,
 //! `BENCH_NAME`, `BENCH_FILTER`, `REPO`, `JOB_ID`, `RUNNER_TOKEN`,
-//! `CONTROLLER_URL`, `RUNNER_REPO_URL`. The scheduled main-tracking
+//! `CONTROLLER_URL`, `RUNNER_REPO_URL`, `SHARED_ENV_VARS`,
+//! `BASELINE_ENV_VARS`, `CHANGED_ENV_VARS`. The scheduled main-tracking
 //! workflow instead supplies `GITHUB_TOKEN` directly (no PR author to
 //! distrust).
 
@@ -67,6 +68,10 @@ pub struct RunnerConfig {
     pub poster_mode: PosterMode,
     pub sccache_gcs_bucket: Option<String>,
     pub data_cache_bucket: Option<String>,
+    /// Env vars from the trigger comment's shared `env:` block. These are
+    /// already set on the pod (so they reach the build too); the map is kept
+    /// only so comments can report what was requested.
+    pub shared_env_vars: HashMap<String, String>,
     pub baseline_env_vars: HashMap<String, String>,
     pub changed_env_vars: HashMap<String, String>,
     pub baseline_ref: Option<String>,
@@ -83,6 +88,10 @@ impl RunnerConfig {
         let comment_url = format!("{pr_url}#issuecomment-{comment_id}");
         let bench_type_str = env_required("BENCH_TYPE")?;
 
+        let shared_env_vars: HashMap<String, String> = std::env::var("SHARED_ENV_VARS")
+            .ok()
+            .and_then(|s| serde_json::from_str(&s).ok())
+            .unwrap_or_default();
         let baseline_env_vars: HashMap<String, String> = std::env::var("BASELINE_ENV_VARS")
             .ok()
             .and_then(|s| serde_json::from_str(&s).ok())
@@ -104,6 +113,7 @@ impl RunnerConfig {
             poster_mode: parse_poster_mode()?,
             sccache_gcs_bucket: std::env::var("SCCACHE_GCS_BUCKET").ok(),
             data_cache_bucket: std::env::var("DATA_CACHE_BUCKET").ok(),
+            shared_env_vars,
             baseline_env_vars,
             changed_env_vars,
             baseline_ref: std::env::var("BASELINE_REF").ok(),

diff --git a/controller/src/runner/mod.rs b/controller/src/runner/mod.rs
@@ -7,3 +7,4 @@ pub mod monitor;
 pub mod pool_peak;
 pub mod poster;
 pub mod shell;
+pub mod trigger;