Merge pull request #236 from knusbaum/knusbaum/troubleshooting

Add instrumentation telemetry endpoint

Author: platinummonkey

docs/COMMANDS.md

@@ -50,7 +50,7 @@ pup <domain> <subgroup> <action> [options] # Nested commands
 | error-tracking | issues (search, get) | src/commands/error_tracking.rs | ✅ |
 | scorecards | list, get | src/commands/scorecards.rs | ✅ |
 | usage | summary, hourly | src/commands/usage.rs | ✅ |
-| apm | services (list, stats, operations, resources), entities (list), dependencies (list), flow-map | src/commands/apm.rs | ✅ |
+| apm | services (list, stats, operations, resources), entities (list), dependencies (list), flow-map, troubleshooting (list) | src/commands/apm.rs | ✅ |
 | cost | projected, attribution, by-org, aws-config (list, get, create, delete), azure-config (list, get, create, delete), gcp-config (list, get, create, delete) | src/commands/cost.rs | ✅ |
 | product-analytics | events send | src/commands/product_analytics.rs | ✅ |
 | data-governance | scanner-rules (list) | src/commands/data_governance.rs | ✅ |

docs/EXAMPLES.md

@@ -285,6 +285,17 @@ pup security findings search \
   --query="@severity:high"
 ```
 
+## APM Troubleshooting
+
+### List Instrumentation Errors for a Host
+```bash
+# Show APM instrumentation errors for a specific host
+pup apm troubleshooting list --hostname my-host
+
+# Narrow results to a specific time window
+pup apm troubleshooting list --hostname my-host --timeframe 4h
+```
+
 ## Infrastructure
 
 ### List Hosts
@@ -296,6 +307,20 @@ pup infrastructure hosts list
 pup infrastructure hosts list --filter="env:production"
 ```
 
+## Fleet
+
+### List Fleet Agents
+```bash
+# Filter agents by hostname
+pup fleet agents list --filter "hostname:my-host"
+
+# Filter by IP address
+pup fleet agents list --filter "ip_address:1.2.3.4"
+
+# Boolean filter expression
+pup fleet agents list --filter "(hostname:host-a OR hostname:host-b) AND env:prod"
+```
+
 ### Get Host
 ```bash
 pup infrastructure hosts get "host-name"

src/commands/apm.rs

@@ -200,3 +200,34 @@ pub async fn flow_map(
     let data = crate::api::get(cfg, "/api/ui/apm/flow-map", &q).await?;
     crate::formatter::output(cfg, &data)
 }
+
+#[cfg(not(target_arch = "wasm32"))]
+pub async fn troubleshooting_list(
+    cfg: &Config,
+    hostname: String,
+    timeframe: Option<String>,
+) -> Result<()> {
+    let path = "/api/unstable/apm/instrumentation-errors";
+    let mut query = vec![("hostname", hostname.as_str())];
+    let tf_owned;
+    if let Some(tf) = &timeframe {
+        tf_owned = tf.clone();
+        query.push(("timeframe", tf_owned.as_str()));
+    }
+    let data = client::raw_get(cfg, path, &query).await?;
+    formatter::output(cfg, &data)
+}
+
+#[cfg(target_arch = "wasm32")]
+pub async fn troubleshooting_list(
+    cfg: &Config,
+    hostname: String,
+    timeframe: Option<String>,
+) -> Result<()> {
+    let mut query = vec![("hostname", hostname)];
+    if let Some(tf) = timeframe {
+        query.push(("timeframe", tf));
+    }
+    let data = crate::api::get(cfg, "/api/unstable/apm/instrumentation-errors", &query).await?;
+    crate::formatter::output(cfg, &data)
+}

src/commands/fleet.rs

@@ -12,7 +12,11 @@ use crate::formatter;
 use crate::util;
 
 #[cfg(not(target_arch = "wasm32"))]
-pub async fn agents_list(cfg: &Config, page_size: Option<i64>) -> Result<()> {
+pub async fn agents_list(
+    cfg: &Config,
+    page_size: Option<i64>,
+    filter: Option<String>,
+) -> Result<()> {
     let dd_cfg = client::make_dd_config(cfg);
     let api = match client::make_bearer_client(cfg) {
         Some(c) => FleetAutomationAPI::with_client_and_config(dd_cfg, c),
@@ -22,6 +26,9 @@ pub async fn agents_list(cfg: &Config, page_size: Option<i64>) -> Result<()> {
     if let Some(ps) = page_size {
         params = params.page_size(ps);
     }
+    if let Some(f) = filter {
+        params = params.filter(f);
+    }
     let resp = api
         .list_fleet_agents(params)
         .await
@@ -30,11 +37,18 @@ pub async fn agents_list(cfg: &Config, page_size: Option<i64>) -> Result<()> {
 }
 
 #[cfg(target_arch = "wasm32")]
-pub async fn agents_list(cfg: &Config, page_size: Option<i64>) -> Result<()> {
+pub async fn agents_list(
+    cfg: &Config,
+    page_size: Option<i64>,
+    filter: Option<String>,
+) -> Result<()> {
     let mut query: Vec<(&str, String)> = Vec::new();
     if let Some(ps) = page_size {
         query.push(("page[size]", ps.to_string()));
     }
+    if let Some(f) = filter {
+        query.push(("filter", f));
+    }
     let data = crate::api::get(cfg, "/api/v2/fleet/agents", &query).await?;
     crate::formatter::output(cfg, &data)
 }

src/main.rs

@@ -4136,6 +4136,11 @@ enum FleetAgentActions {
     List {
         #[arg(long)]
         page_size: Option<i64>,
+        #[arg(
+            long,
+            help = "Filter query (e.g. ip_address:1.2.3.4, hostname:my-host)"
+        )]
+        filter: Option<String>,
     },
     /// Get fleet agent details
     Get { agent_key: String },
@@ -4756,6 +4761,11 @@ enum ApmActions {
         #[arg(long, help = "Environment filter")]
         env: Option<String>,
     },
+    /// Troubleshoot APM instrumentation issues
+    Troubleshooting {
+        #[command(subcommand)]
+        action: ApmTroubleshootingActions,
+    },
 }
 
 #[derive(Subcommand)]
@@ -4854,6 +4864,17 @@ enum ApmDependencyActions {
     },
 }
 
+#[derive(Subcommand)]
+enum ApmTroubleshootingActions {
+    /// List instrumentation errors for a host
+    List {
+        #[arg(long, help = "Hostname to query (required)")]
+        hostname: String,
+        #[arg(long, help = "Time window (e.g. 4h, 24h, 1h30m)")]
+        timeframe: Option<String>,
+    },
+}
+
 // ---- Investigations ----
 #[derive(Subcommand)]
 enum InvestigationActions {
@@ -7400,8 +7421,8 @@ async fn main_inner() -> anyhow::Result<()> {
             cfg.validate_auth()?;
             match action {
                 FleetActions::Agents { action } => match action {
-                    FleetAgentActions::List { page_size } => {
-                        commands::fleet::agents_list(&cfg, page_size).await?;
+                    FleetAgentActions::List { page_size, filter } => {
+                        commands::fleet::agents_list(&cfg, page_size, filter).await?;
                     }
                     FleetAgentActions::Get { agent_key } => {
                         commands::fleet::agents_get(&cfg, &agent_key).await?;
@@ -7894,6 +7915,14 @@ async fn main_inner() -> anyhow::Result<()> {
                 } => {
                     commands::apm::flow_map(&cfg, query, limit, from, to).await?;
                 }
+                ApmActions::Troubleshooting { action } => match action {
+                    ApmTroubleshootingActions::List {
+                        hostname,
+                        timeframe,
+                    } => {
+                        commands::apm::troubleshooting_list(&cfg, hostname, timeframe).await?;
+                    }
+                },
             }
         }
         // --- DDSQL ---

src/test_commands.rs

@@ -1427,7 +1427,7 @@ async fn test_fleet_agents_list() {
     let mut s = mockito::Server::new_async().await;
     let cfg = test_config(&s.url());
     mock_all(&mut s, r#"{"data": []}"#).await;
-    let _ = crate::commands::fleet::agents_list(&cfg, None).await;
+    let _ = crate::commands::fleet::agents_list(&cfg, None, None).await;
     cleanup_env();
 }
 #[tokio::test]
@@ -2054,6 +2054,61 @@ async fn test_apm_services_list() {
         crate::commands::apm::services_list(&cfg, "prod".into(), "1h".into(), "now".into()).await;
     cleanup_env();
 }
+#[tokio::test]
+async fn test_apm_troubleshooting_list() {
+    let _lock = lock_env();
+    let mut server = mockito::Server::new_async().await;
+    let cfg = test_config(&server.url());
+
+    let mock = server
+        .mock("GET", "/api/unstable/apm/instrumentation-errors")
+        .match_query(mockito::Matcher::UrlEncoded(
+            "hostname".into(),
+            "my-host".into(),
+        ))
+        .with_status(200)
+        .with_header("content-type", "application/json")
+        .with_body(r#"{"data": []}"#)
+        .create_async()
+        .await;
+
+    let result = crate::commands::apm::troubleshooting_list(&cfg, "my-host".into(), None).await;
+    assert!(
+        result.is_ok(),
+        "troubleshooting list failed: {:?}",
+        result.err()
+    );
+    mock.assert_async().await;
+    cleanup_env();
+}
+#[tokio::test]
+async fn test_apm_troubleshooting_list_with_timeframe() {
+    let _lock = lock_env();
+    let mut server = mockito::Server::new_async().await;
+    let cfg = test_config(&server.url());
+
+    let mock = server
+        .mock("GET", "/api/unstable/apm/instrumentation-errors")
+        .match_query(mockito::Matcher::AllOf(vec![
+            mockito::Matcher::UrlEncoded("hostname".into(), "my-host".into()),
+            mockito::Matcher::UrlEncoded("timeframe".into(), "4h".into()),
+        ]))
+        .with_status(200)
+        .with_header("content-type", "application/json")
+        .with_body(r#"{"data": []}"#)
+        .create_async()
+        .await;
+
+    let result =
+        crate::commands::apm::troubleshooting_list(&cfg, "my-host".into(), Some("4h".into())).await;
+    assert!(
+        result.is_ok(),
+        "troubleshooting list with timeframe failed: {:?}",
+        result.err()
+    );
+    mock.assert_async().await;
+    cleanup_env();
+}
 
 // -------------------------------------------------------------------------
 // Read-only mode