observability: Add configurable log file retention limit

Add a positive log_file_retention_limit config value with a default of 10 and expose its resolved provenance through config show.

Apply the resolved limit to retention cleanup for primary and v2 log files.

Co-authored-by: SCE <sce@crocoder.dev>

Author: 2 people

cli/assets/generated/config/schema/sce-config.schema.json

@@ -29,6 +29,11 @@
       "type": "string",
       "minLength": 1
     },
+    "log_file_retention_limit": {
+      "default": 10,
+      "type": "integer",
+      "minimum": 1
+    },
     "timeout_ms": {
       "type": "integer",
       "minimum": 0

cli/src/services/config/render.rs

@@ -67,6 +67,10 @@ pub(super) fn format_show_output(runtime: &RuntimeConfig, report_format: ReportF
                             runtime.log_format.source,
                         ),
                         "log_dir": format_optional_resolved_value_json(&runtime.log_dir),
+                        "log_file_retention_limit": format_resolved_value_json(
+                            runtime.log_file_retention_limit.value,
+                            runtime.log_file_retention_limit.source,
+                        ),
                         "timeout_ms": {
                             "value": runtime.timeout_ms.value,
                             "source": runtime.timeout_ms.source.as_str(),
@@ -217,6 +221,11 @@ fn format_observability_text_lines(runtime: &RuntimeConfig) -> Vec<String> {
             runtime.log_format.source,
         ),
         format_optional_resolved_value_text("log_dir", &runtime.log_dir),
+        format_resolved_value_text(
+            "log_file_retention_limit",
+            &runtime.log_file_retention_limit.value.to_string(),
+            runtime.log_file_retention_limit.source,
+        ),
     ]
 }
 

cli/src/services/config/resolver.rs

@@ -18,8 +18,8 @@ use super::types::{
     parse_bool_value_from, ConfigPathSource, ConfigRequest, DatabaseRetryConfig, LoadedConfigPath,
     LogFormat, LogLevel, ReportFormat, ResolvedAgentTraceStorageRuntimeConfig,
     ResolvedAuthRuntimeConfig, ResolvedHookRuntimeConfig, ResolvedObservabilityRuntimeConfig,
-    ResolvedOptionalValue, ResolvedValue, ValueSource, ENV_ATTRIBUTION_HOOKS_DISABLED, ENV_LOG_DIR,
-    ENV_LOG_FORMAT, ENV_LOG_LEVEL,
+    ResolvedOptionalValue, ResolvedValue, ValueSource, DEFAULT_LOG_FILE_RETENTION_LIMIT,
+    ENV_ATTRIBUTION_HOOKS_DISABLED, ENV_LOG_DIR, ENV_LOG_FORMAT, ENV_LOG_LEVEL,
 };
 
 const DEFAULT_TIMEOUT_MS: u64 = 30000;
@@ -62,6 +62,7 @@ pub(super) struct RuntimeConfig {
     pub(super) log_level: ResolvedValue<LogLevel>,
     pub(super) log_format: ResolvedValue<LogFormat>,
     pub(super) log_dir: ResolvedOptionalValue<String>,
+    pub(super) log_file_retention_limit: ResolvedValue<usize>,
     pub(super) timeout_ms: ResolvedValue<u64>,
     pub(super) attribution_hooks_enabled: ResolvedValue<bool>,
     pub(super) workos_client_id: ResolvedOptionalValue<String>,
@@ -222,6 +223,7 @@ where
         log_level: runtime.log_level.value,
         log_format: runtime.log_format.value,
         log_dir: runtime.log_dir.value,
+        log_file_retention_limit: runtime.log_file_retention_limit.value,
         loaded_config_paths: runtime.loaded_config_paths,
         validation_errors: runtime.validation_errors,
     })
@@ -298,6 +300,7 @@ where
         log_level: None,
         log_format: None,
         log_dir: None,
+        log_file_retention_limit: None,
         timeout_ms: None,
         attribution_hooks_enabled: None,
         workos_client_id: None,
@@ -328,6 +331,9 @@ where
         if let Some(log_dir) = layer.log_dir {
             file_config.log_dir = Some(log_dir);
         }
+        if let Some(log_file_retention_limit) = layer.log_file_retention_limit {
+            file_config.log_file_retention_limit = Some(log_file_retention_limit);
+        }
         if let Some(timeout_ms) = layer.timeout_ms {
             file_config.timeout_ms = Some(timeout_ms);
         }
@@ -411,6 +417,17 @@ where
         default_observability_log_dir()?
     };
 
+    let resolved_log_file_retention_limit = match file_config.log_file_retention_limit {
+        Some(value) => ResolvedValue {
+            value: value.value,
+            source: ValueSource::ConfigFile(value.source),
+        },
+        None => ResolvedValue {
+            value: DEFAULT_LOG_FILE_RETENTION_LIMIT,
+            source: ValueSource::Default,
+        },
+    };
+
     let mut resolved_timeout_ms = ResolvedValue {
         value: DEFAULT_TIMEOUT_MS,
         source: ValueSource::Default,
@@ -499,6 +516,7 @@ where
         log_level: resolved_log_level,
         log_format: resolved_log_format,
         log_dir: resolved_log_dir,
+        log_file_retention_limit: resolved_log_file_retention_limit,
         timeout_ms: resolved_timeout_ms,
         attribution_hooks_enabled: resolved_attribution_hooks_enabled,
         workos_client_id: resolved_workos_client_id,

cli/src/services/config/schema.rs

@@ -33,6 +33,7 @@ pub(crate) const TOP_LEVEL_CONFIG_KEYS: &[&str] = &[
     "log_level",
     "log_format",
     "log_dir",
+    "log_file_retention_limit",
     "timeout_ms",
     super::resolver::WORKOS_CLIENT_ID_KEY.config_key,
     "agent_trace",
@@ -41,7 +42,7 @@ pub(crate) const TOP_LEVEL_CONFIG_KEYS: &[&str] = &[
 ];
 
 pub(crate) const TOP_LEVEL_CONFIG_KEYS_DESCRIPTION: &str =
-    "$schema, log_level, log_format, timeout_ms, workos_client_id, agent_trace, policies, integrations, log_dir";
+    "$schema, log_level, log_format, timeout_ms, workos_client_id, agent_trace, policies, integrations, log_dir, log_file_retention_limit";
 
 static CONFIG_SCHEMA_VALIDATOR: OnceLock<Validator> = OnceLock::new();
 
@@ -68,6 +69,7 @@ pub(crate) struct ParsedFileConfigDocument {
     pub(crate) log_level: Option<String>,
     pub(crate) log_format: Option<String>,
     pub(crate) log_dir: Option<String>,
+    pub(crate) log_file_retention_limit: Option<usize>,
     pub(crate) timeout_ms: Option<u64>,
     pub(crate) workos_client_id: Option<String>,
     pub(crate) agent_trace: Option<ParsedAgentTraceConfigDocument>,
@@ -151,6 +153,7 @@ pub(crate) struct FileConfig {
     pub(crate) log_level: Option<FileConfigValue<LogLevel>>,
     pub(crate) log_format: Option<FileConfigValue<LogFormat>>,
     pub(crate) log_dir: Option<FileConfigValue<String>>,
+    pub(crate) log_file_retention_limit: Option<FileConfigValue<usize>>,
     pub(crate) timeout_ms: Option<FileConfigValue<u64>>,
     pub(crate) attribution_hooks_enabled: Option<FileConfigValue<bool>>,
     pub(crate) workos_client_id: Option<FileConfigValue<String>>,
@@ -285,6 +288,9 @@ pub(crate) fn parse_file_config(
         })
         .transpose()?;
     let log_dir = typed.log_dir.map(|value| FileConfigValue { value, source });
+    let log_file_retention_limit = typed
+        .log_file_retention_limit
+        .map(|value| FileConfigValue { value, source });
     let timeout_ms = typed
         .timeout_ms
         .map(|value| FileConfigValue { value, source });
@@ -301,6 +307,7 @@ pub(crate) fn parse_file_config(
         log_level,
         log_format,
         log_dir,
+        log_file_retention_limit,
         timeout_ms,
         attribution_hooks_enabled,
         workos_client_id,

cli/src/services/config/types.rs

@@ -17,6 +17,7 @@ pub(crate) const ENV_LOG_LEVEL: &str = "SCE_LOG_LEVEL";
 pub(crate) const ENV_LOG_FORMAT: &str = "SCE_LOG_FORMAT";
 pub(crate) const ENV_LOG_DIR: &str = "SCE_LOG_DIR";
 pub(crate) const ENV_ATTRIBUTION_HOOKS_DISABLED: &str = "SCE_ATTRIBUTION_HOOKS_DISABLED";
+pub(crate) const DEFAULT_LOG_FILE_RETENTION_LIMIT: usize = 10;
 
 pub type ReportFormat = OutputFormat;
 
@@ -197,6 +198,7 @@ pub(crate) struct ResolvedObservabilityRuntimeConfig {
     pub(crate) log_level: LogLevel,
     pub(crate) log_format: LogFormat,
     pub(crate) log_dir: Option<String>,
+    pub(crate) log_file_retention_limit: usize,
     pub(crate) loaded_config_paths: Vec<LoadedConfigPath>,
     pub(crate) validation_errors: Vec<String>,
 }

cli/src/services/observability.rs

@@ -26,7 +26,6 @@ pub const NAME: &str = "observability";
 const LOG_FILE_PREFIX: &str = "sce";
 const LOG_FILE_EXTENSION: &str = "log";
 const EMPTY_SESSION_ID_TOKEN: &str = "%EMPTY";
-const LOG_FILE_RETENTION_LIMIT: usize = 10;
 
 #[derive(Clone, Copy, Debug, Eq, PartialEq)]
 pub struct ObservabilityConfig {
@@ -47,6 +46,7 @@ impl Default for ObservabilityConfig {
 pub struct Logger {
     config: ObservabilityConfig,
     log_dir: Option<PathBuf>,
+    log_file_retention_limit: usize,
 }
 
 impl Logger {
@@ -63,6 +63,7 @@ impl Logger {
                 format: config.log_format,
             },
             log_dir: config.log_dir.as_deref().map(PathBuf::from),
+            log_file_retention_limit: config.log_file_retention_limit,
         })
     }
 
@@ -87,7 +88,11 @@ impl Logger {
             log_dir = Some(PathBuf::from(raw));
         }
 
-        Ok(Self { config, log_dir })
+        Ok(Self {
+            config,
+            log_dir,
+            log_file_retention_limit: config::DEFAULT_LOG_FILE_RETENTION_LIMIT,
+        })
     }
 
     pub fn info(
@@ -188,7 +193,7 @@ impl Logger {
         };
 
         let path = current_log_path(log_dir, session_id);
-        append_log_line(&path, redacted_line)
+        append_log_line(&path, redacted_line, self.log_file_retention_limit)
     }
 
     fn enabled(&self, level: LogLevel) -> bool {
@@ -294,8 +299,10 @@ fn sanitize_session_id_for_filename(session_id: &str) -> String {
     sanitized
 }
 
-fn append_log_line(path: &Path, redacted_line: &str) -> Result<()> {
-    append_log_line_with_cleanup(path, redacted_line, enforce_log_retention)
+fn append_log_line(path: &Path, redacted_line: &str, retention_limit: usize) -> Result<()> {
+    append_log_line_with_cleanup(path, redacted_line, |log_dir| {
+        enforce_log_retention(log_dir, retention_limit)
+    })
 }
 
 fn append_log_line_with_cleanup<F>(path: &Path, redacted_line: &str, cleanup: F) -> Result<()>
@@ -440,11 +447,15 @@ struct ManagedLogFile {
     modified: SystemTime,
 }
 
-fn enforce_log_retention(log_dir: &Path) -> Result<()> {
-    enforce_log_retention_with(log_dir, |path| fs::remove_file(path))
+fn enforce_log_retention(log_dir: &Path, retention_limit: usize) -> Result<()> {
+    enforce_log_retention_with(log_dir, retention_limit, |path| fs::remove_file(path))
 }
 
-fn enforce_log_retention_with<F>(log_dir: &Path, mut remove_file: F) -> Result<()>
+fn enforce_log_retention_with<F>(
+    log_dir: &Path,
+    retention_limit: usize,
+    mut remove_file: F,
+) -> Result<()>
 where
     F: FnMut(&Path) -> io::Result<()>,
 {
@@ -455,7 +466,7 @@ where
         ordering => ordering,
     });
 
-    for managed_file in managed_files.into_iter().skip(LOG_FILE_RETENTION_LIMIT) {
+    for managed_file in managed_files.into_iter().skip(retention_limit) {
         if let Err(error) = remove_file(&managed_file.path) {
             errors.push(format!(
                 "failed to remove old log file '{}': {error}",

config/pkl/base/sce-config-schema.pkl

@@ -74,6 +74,11 @@ local sceConfigSchema = new JsonSchema {
       type = "string"
       minLength = 1
     }
+    ["log_file_retention_limit"] = new JsonSchema {
+      type = "integer"
+      minimum = 1
+      default = 10
+    }
     ["timeout_ms"] = new JsonSchema {
       type = "integer"
       minimum = 0

config/schema/sce-config.schema.json

@@ -29,6 +29,11 @@
       "type": "string",
       "minLength": 1
     },
+    "log_file_retention_limit": {
+      "default": 10,
+      "type": "integer",
+      "minimum": 1
+    },
     "timeout_ms": {
       "type": "integer",
       "minimum": 0