rails_error_dashboard 0.12.1 → 0.14.0

data/config/locales/ru.yml

@@ -441,6 +441,7 @@ ru:
       issue_section:
         title: Трекер задач
         view_issue: Открыть задачу
+        unsafe_link: "Эта ссылка на задачу не является URL-адресом http(s), поэтому она показана как текст и не может быть открыта."
         linked: Привязана
         issue_reference: "%{provider} #%{number}"
         states:
@@ -613,6 +614,7 @@ ru:
         snapshot_stale: из более раннего появления — последующие не содержали контекста
         snapshot_fidelity_lite: сокращённый сбор (защита от storm отбросила payload контекста)
         snapshot_fidelity_minimal: учтено во время storm — контекст не собирался
+        snapshot_fidelity_partial: смешанный снимок — часть полей осталась от более раннего события
         snapshot_unknown: получен до того, как RED начал записывать происхождение снимков
       similar_errors:
         title: Похожие ошибки
@@ -980,6 +982,7 @@ ru:
         rate_per_hour: "%{value}/ч"
         affected_users_incomplete: не менее — защита от storm отбросила детализацию по событиям
         data_unavailable: Статистика сейчас недоступна — приведённые ниже числа не являются измерением.
+        event_timing_incomplete: "Отметки времени неполные — защита от шторма отбросила их; итоговые значения остаются точными."
         unresolved: Не решено
         unresolved_hint: Ожидают решения
         resolution_rate: Доля решённых
@@ -1019,6 +1022,16 @@ ru:
           other: "%{count} типа"
           few: "%{count} типа"
           many: "%{count} типов"
+    errors_page:
+      csrf:
+        title: "Срок действия токена сеанса истёк"
+        message: "Вернитесь назад, перезагрузите страницу и повторите попытку. Ничего не было изменено."
+      bad_request:
+        title: "Не удалось обработать этот запрос"
+        message: "Обязательное значение отсутствовало или имело неверный формат. Вернитесь назад и повторите попытку."
+      not_acceptable:
+        title: "Этот формат недоступен"
+        message: "Панель не может сформировать формат, запрошенный в этом запросе."
     settings:
       title: Настройки
       version: v%{version}
@@ -1151,6 +1164,8 @@ ru:
         application_name: Имя этого приложения (определяется автоматически, если не
           задано)
         environment: Окружение, к которому относятся ошибки (Rails.env, если не задано)
+        notification_burst_limit: "Уведомления о новых ошибках за интервал на процесс (0 = без ограничения)"
+        notification_burst_window_seconds: "Длительность интервала уведомлений о новых ошибках"
         notification_environments: Уведомлять только для этих окружений (все, если
           не задано)
         database: Активная база данных для журнала ошибок
@@ -1261,6 +1276,16 @@ ru:
         linked: Задача успешно привязана
         create_failed: 'Не удалось создать задачу: %{reason}'
         link_failed: 'Не удалось привязать задачу: %{reason}'
+      status:
+        updated: "Статус изменён на %{status}"
+        invalid_transition: "Нельзя изменить статус с %{from} на %{to}"
+        unknown: "Такого статуса не существует"
+      snooze:
+        invalid_hours: "Время откладывания должно быть целым числом часов от 1 до %{max}"
+      priority:
+        invalid: "Такого уровня приоритета не существует"
+      assign:
+        blank: "Введите имя того, кому назначить эту ошибку"
       not_enabled: "%{feature} не включено. Включите в %{file}"
       not_enabled_plural: "%{feature} не включены. Включите их в %{file}"
       not_enabled_options: "%{feature} не включено. Включите %{options} в %{file}"
@@ -1303,6 +1328,7 @@ ru:
         not_configured: Отслеживание задач не настроено
         already_linked: 'К ошибке уже привязана задача: %{url}'
         url_required: Требуется URL задачи
+        url_invalid: URL задачи должен начинаться с http:// или https://
       error_not_found: 'Ошибка не найдена: %{id}'
       batch_failed:
         mute:
@@ -1441,6 +1467,10 @@ ru:
           few: "%{value} ошибки"
           many: "%{value} ошибок"
         view_in_dashboard: Открыть в панели
+      burst:
+        suppressed: "В приложении %{application} за %{window} секунд зафиксировано более %{limit} новых ошибок."
+        still_recorded: "Уведомления о дальнейших новых ошибках приостановлены до конца этого интервала; каждая ошибка по-прежнему записывается."
+        dashboard: "Панель: %{url}"
       storm:
         detected: Обнаружен шторм ошибок в %{application} в %{started_at}.
         mode:

data/config/locales/uk.yml

@@ -438,6 +438,7 @@ uk:
       issue_section:
         title: Трекер задач
         view_issue: Переглянути задачу
+        unsafe_link: "Це посилання на задачу не є URL-адресою http(s), тому воно показане як текст і не може бути відкрите."
         linked: Приєднано
         issue_reference: "%{provider} #%{number}"
         states:
@@ -611,6 +612,7 @@ uk:
         snapshot_stale: з ранішого випадку — подальші не містили контексту
         snapshot_fidelity_lite: скорочений збір (захист від storm відкинув payload контексту)
         snapshot_fidelity_minimal: враховано під час storm — контекст не збирався
+        snapshot_fidelity_partial: змішаний знімок — частина полів залишилася від давнішої події
         snapshot_unknown: отримано до того, як RED почав записувати походження знімків
       similar_errors:
         title: Схожі помилки
@@ -977,6 +979,7 @@ uk:
         rate_per_hour: "%{value}/год"
         affected_users_incomplete: щонайменше — захист від storm відкинув деталізацію за подіями
         data_unavailable: Статистика зараз недоступна — наведені нижче числа не є вимірюванням.
+        event_timing_incomplete: "Позначки часу неповні — захист від шторму відкинув їх; підсумкові значення залишаються точними."
         unresolved: Не вирішено
         unresolved_hint: Очікують вирішення
         resolution_rate: Частка вирішених
@@ -1016,6 +1019,16 @@ uk:
           other: "%{count} типу"
           few: "%{count} типи"
           many: "%{count} типів"
+    errors_page:
+      csrf:
+        title: "Термін дії токена сеансу минув"
+        message: "Поверніться назад, перезавантажте сторінку та спробуйте ще раз. Нічого не було змінено."
+      bad_request:
+        title: "Не вдалося обробити цей запит"
+        message: "Обов'язкове значення було відсутнє або мало неправильний формат. Поверніться назад і спробуйте ще раз."
+      not_acceptable:
+        title: "Цей формат недоступний"
+        message: "Панель не може сформувати формат, запитаний у цьому запиті."
     settings:
       title: Налаштування
       version: v%{version}
@@ -1147,6 +1160,8 @@ uk:
         application_name: Назва цього застосунку (визначається автоматично, якщо не
           задано)
         environment: Середовище, до якого належать помилки (Rails.env, якщо не задано)
+        notification_burst_limit: "Сповіщення про нові помилки за інтервал на процес (0 = без обмеження)"
+        notification_burst_window_seconds: "Тривалість інтервалу сповіщень про нові помилки"
         notification_environments: Сповіщати лише для цих середовищ (усі, якщо не
           задано)
         database: Активна база даних для журналів помилок
@@ -1260,6 +1275,16 @@ uk:
         linked: Задачу успішно приєднано
         create_failed: 'Не вдалося створити задачу: %{reason}'
         link_failed: 'Не вдалося приєднати задачу: %{reason}'
+      status:
+        updated: "Статус змінено на %{status}"
+        invalid_transition: "Неможливо змінити статус з %{from} на %{to}"
+        unknown: "Такого статусу не існує"
+      snooze:
+        invalid_hours: "Час відкладення має бути цілим числом годин від 1 до %{max}"
+      priority:
+        invalid: "Такого рівня пріоритету не існує"
+      assign:
+        blank: "Введіть ім'я того, кому призначити цю помилку"
       not_enabled: "%{feature} не увімкнено. Увімкніть у %{file}"
       not_enabled_plural: "%{feature} не увімкнено. Увімкніть їх у %{file}"
       not_enabled_options: "%{feature} не увімкнено. Увімкніть %{options} у %{file}"
@@ -1301,6 +1326,7 @@ uk:
         not_configured: Відстеження задач не налаштовано
         already_linked: 'До помилки вже приєднано задачу: %{url}'
         url_required: Потрібна URL-адреса задачі
+        url_invalid: URL-адреса задачі має починатися з http:// або https://
       error_not_found: 'Помилку не знайдено: %{id}'
       batch_failed:
         mute:
@@ -1439,6 +1465,10 @@ uk:
           few: "%{value} помилки"
           many: "%{value} помилок"
         view_in_dashboard: Переглянути в панелі
+      burst:
+        suppressed: "У застосунку %{application} за %{window} секунд зафіксовано понад %{limit} нових помилок."
+        still_recorded: "Сповіщення про подальші нові помилки призупинено до кінця цього інтервалу; кожна помилка, як і раніше, записується."
+        dashboard: "Панель: %{url}"
       storm:
         detected: Виявлено шторм помилок у %{application} о %{started_at}.
         mode:

data/config/locales/zh-CN.yml

@@ -380,6 +380,7 @@ zh-CN:
       issue_section:
         title: 工单系统
         view_issue: 查看工单
+        unsafe_link: "此工单链接不是 http(s) URL，因此仅以文本显示，无法打开。"
         linked: 已关联
         issue_reference: "%{provider} #%{number}"
         states:
@@ -528,6 +529,7 @@ zh-CN:
         snapshot_stale: 来自更早的一次发生 — 之后的发生未携带上下文
         snapshot_fidelity_lite: 已精简的捕获（storm 保护丢弃了上下文 payload）
         snapshot_fidelity_minimal: 在 storm 期间计数 — 未捕获任何上下文
+        snapshot_fidelity_partial: 混合快照 — 部分字段来自更早的一次发生
         snapshot_unknown: 在 RED 记录快照来源之前捕获
       similar_errors:
         title: 相似错误
@@ -841,6 +843,7 @@ zh-CN:
         rate_per_hour: "%{value}/小时"
         affected_users_incomplete: 至少 — storm 保护丢弃了逐事件的明细
         data_unavailable: 统计数据当前不可用 — 下方数字并非测量结果。
+        event_timing_incomplete: "事件时间记录不完整 — 风暴保护丢弃了时间戳；总数仍然准确。"
         unresolved: 未解决
         unresolved_hint: 等待处理
         resolution_rate: 解决率
@@ -870,6 +873,16 @@ zh-CN:
         user_types_html: 用户 %{user} · %{types}
         types:
           other: "%{count} 种类型"
+    errors_page:
+      csrf:
+        title: "会话令牌已过期"
+        message: "请返回并重新加载页面后重试。未做任何更改。"
+      bad_request:
+        title: "无法理解此请求"
+        message: "缺少必需的值或其格式不正确。请返回后重试。"
+      not_acceptable:
+        title: "此格式不可用"
+        message: "仪表板无法生成此请求所要求的格式。"
     settings:
       title: 设置
       version: v%{version}
@@ -963,6 +976,8 @@ zh-CN:
         sampling_rate: 错误捕获的采样比例
         application_name: 本应用的名称（未设置时自动检测）
         environment: 错误所属的环境（未设置时为 Rails.env）
+        notification_burst_limit: "每个时间窗口、每个进程的新错误通知数（0 = 不限制）"
+        notification_burst_window_seconds: "新错误通知时间窗口的长度"
         notification_environments: 仅对这些环境发送通知（未设置时为全部）
         database: 错误日志使用的数据库
         use_separate_database: 为错误日志使用独立数据库
@@ -1049,6 +1064,16 @@ zh-CN:
         linked: 工单关联成功
         create_failed: 创建工单失败：%{reason}
         link_failed: 关联工单失败：%{reason}
+      status:
+        updated: "状态已更改为 %{status}"
+        invalid_transition: "无法将状态从 %{from} 更改为 %{to}"
+        unknown: "该状态不存在"
+      snooze:
+        invalid_hours: "延后时间必须是 1 到 %{max} 之间的整数小时"
+      priority:
+        invalid: "该优先级不存在"
+      assign:
+        blank: "请输入要分配此错误的人员姓名"
       not_enabled: "%{feature}未启用。请在 %{file} 中启用。"
       not_enabled_plural: "%{feature}未启用。请在 %{file} 中启用。"
       not_enabled_options: "%{feature}未启用。请在 %{file} 中启用 %{options}。"
@@ -1087,6 +1112,7 @@ zh-CN:
         not_configured: 工单跟踪尚未配置
         already_linked: 该错误已关联工单：%{url}
         url_required: 必须提供工单 URL
+        url_invalid: 工单 URL 必须以 http:// 或 https:// 开头
       error_not_found: 未找到错误：%{id}
       batch_failed:
         mute:
@@ -1203,6 +1229,10 @@ zh-CN:
         threshold_errors:
           other: "%{value} 个错误"
         view_in_dashboard: 在面板中查看
+      burst:
+        suppressed: "%{application} 在 %{window} 秒内报告了超过 %{limit} 个新错误。"
+        still_recorded: "在该时间窗口结束之前，后续新错误的通知将暂停；所有错误仍会被记录。"
+        dashboard: "面板：%{url}"
       storm:
         detected: 在 %{application} 于 %{started_at} 检测到错误风暴。
         mode:

data/db/migrate/20260917000001_add_last_notified_at_to_error_logs.rb

@@ -0,0 +1,40 @@
+# frozen_string_literal: true
+
+# When a notification was last sent for this error group.
+#
+# The notification cooldown used to live in a Hash inside each process. Every
+# Puma worker and every job process had its own copy, so one bad deploy that
+# reopened an error notified once PER PROCESS, and a restart forgot the
+# cooldown entirely.
+#
+# With the timestamp on the row the cooldown is claimed in the database with a
+# single conditional UPDATE, which exactly one process can win:
+#
+#   UPDATE ... SET last_notified_at = now
+#   WHERE id = ? AND (last_notified_at IS NULL OR last_notified_at < cutoff)
+#
+# No index: the row is always addressed by primary key.
+#
+# Nullable, no default: NULL means "never notified". Until this migration has
+# run the gem falls back to the in-process cooldown, so upgrading the gem
+# before migrating is safe.
+#
+# up/down rather than `change`: the column_exists? guard that makes `up` safe
+# to re-run would make a reversed `change` skip the removal.
+class AddLastNotifiedAtToErrorLogs < ActiveRecord::Migration[7.0]
+  TABLE = :rails_error_dashboard_error_logs
+
+  def up
+    return unless table_exists?(TABLE)
+    return if column_exists?(TABLE, :last_notified_at)
+
+    add_column TABLE, :last_notified_at, :datetime
+  end
+
+  def down
+    return unless table_exists?(TABLE)
+    return unless column_exists?(TABLE, :last_notified_at)
+
+    remove_column TABLE, :last_notified_at
+  end
+end

data/db/migrate/20260919000001_create_event_counts.rb

@@ -0,0 +1,71 @@
+# frozen_string_literal: true
+
+# Time buckets for storm-shed event volume.
+#
+# Every time-window figure on the dashboard ("errors today", the daily trend,
+# the hourly chart, the error rate) used to filter error_logs by occurred_at
+# and then SUM(occurrence_count). occurred_at is FIRST-SEEN and is never
+# rewritten on recurrence, so that sum reports the lifetime volume of the
+# groups born inside the window -- not the events that happened in it. An
+# error first seen at 23:59 and recurring at 00:01 reported "0 errors today"
+# and put both events on yesterday.
+#
+# The obvious fix -- count error_occurrences rows in the window -- is wrong on
+# its own. Storm protection deliberately writes NO occurrence row while
+# shedding: it folds N events into occurrence_count and nothing else. Counting
+# occurrence rows would therefore erase storm volume from every time window,
+# which is exactly when the numbers matter most.
+#
+# So shed volume needs its own timestamp, and this is it: one row per
+# (error group, 15-minute bucket), holding how many shed events landed in it.
+# Window volume is then
+#
+#   occurrence rows in the window + shed bucket counts in the window
+#
+# which is correct for both ordinary and shed events.
+#
+# Small by construction: a row is written only while a storm is actually
+# shedding, at most one per group per quarter hour.
+#
+# Cleanup follows the GROUP, not the bucket's own age: RetentionCleanupJob
+# deletes the buckets of groups it expires, and ErrorLog has_many :event_counts
+# with dependent: :delete_all covers an explicit destroy. Buckets of a
+# still-active group are deliberately kept -- pruning them by bucket_at alone
+# would silently redistribute those events onto the group's first-seen date,
+# because EventVolume falls back to the group's lifetime count for any group
+# with no per-event record left.
+#
+# (An earlier version of this comment claimed pruning "on bucket_at" that was
+# never implemented. Both mechanisms above are now covered by
+# spec/models/rails_error_dashboard/event_count_cleanup_spec.rb.)
+class CreateEventCounts < ActiveRecord::Migration[7.0]
+  def change
+    # Guard against a squashed schema migration having already created this
+    # table -- without it, every later migration is silently cancelled.
+    return if table_exists?(:rails_error_dashboard_event_counts)
+
+    create_table :rails_error_dashboard_event_counts do |t|
+      t.bigint :error_log_id, null: false
+      # Truncated to a 15-MINUTE bucket, in UTC (EventCount::BUCKET_SECONDS,
+      # shared with the producer). Not the hour: every UTC offset in use
+      # divides into 15 minutes -- including +05:30 and +05:45 -- so a local
+      # midnight falls on a bucket EDGE and a day's total is exact. The row
+      # count stays bounded even through a long storm: at most one row per
+      # group per quarter hour, and only while shedding.
+      t.datetime :bucket_at, null: false
+      t.bigint :count, null: false, default: 0
+      t.timestamps
+    end
+
+    # The upsert target: one bucket per group per quarter hour. Named explicitly --
+    # an auto-generated name here would exceed PostgreSQL's 63-character limit
+    # and fail during the HOST app's deploy (see mailboxer#480).
+    add_index :rails_error_dashboard_event_counts, [ :error_log_id, :bucket_at ],
+              unique: true,
+              name: "index_red_event_counts_on_group_and_bucket"
+
+    # Window queries scan by time; retention prunes on the same column.
+    add_index :rails_error_dashboard_event_counts, :bucket_at,
+              name: "index_red_event_counts_on_bucket_at"
+  end
+end

data/db/migrate/20260920000001_add_buckets_incomplete_to_storm_events.rb

@@ -0,0 +1,25 @@
+# frozen_string_literal: true
+
+# Records that a storm episode lost its per-bucket TIMING evidence.
+#
+# When the rollup table is permanently unavailable (never migrated, or an
+# adapter that refuses the statement), FlushStormCounts deliberately degrades:
+# it keeps the authoritative lifetime occurrence_count and skips the bucket.
+# That is the right trade -- losing a time bucket beats losing a count -- but
+# it leaves every time-window figure for that episode resting on the group's
+# own occurred_at rather than on per-event evidence.
+#
+# The flag was returned in the flush result and then dropped on the floor: no
+# caller persisted it, no query read it, and nothing on the dashboard said so.
+# A replay lost it entirely. Persisting it on the episode is what lets the
+# Overview report an incomplete dimension instead of presenting a figure of
+# unknown completeness as fact -- exactly as affected_users_incomplete already
+# does for storm-shed occurrence rows.
+class AddBucketsIncompleteToStormEvents < ActiveRecord::Migration[7.0]
+  def change
+    return unless table_exists?(:rails_error_dashboard_storm_events)
+    return if column_exists?(:rails_error_dashboard_storm_events, :buckets_incomplete)
+
+    add_column :rails_error_dashboard_storm_events, :buckets_incomplete, :boolean, default: false
+  end
+end

data/db/migrate/20260920000002_create_event_timing_gaps.rb

@@ -0,0 +1,55 @@
+# frozen_string_literal: true
+
+# One row per interval whose per-event TIMING evidence was lost.
+#
+# When the rollup table is permanently unavailable, FlushStormCounts keeps the
+# authoritative lifetime occurrence_count and skips the bucket. The counts stay
+# exact; only their placement in time becomes approximate. The dashboard has to
+# say so, or it presents a figure of unknown completeness as fact.
+#
+# Three earlier attempts to carry that fact failed, each for a structural
+# reason, and this table exists to fix all three at once:
+#
+#   1. A boolean in the command's RETURN HASH. A background job's return value
+#      never reaches the dashboard.
+#   2. A flag on the storm EPISODE. The episode is OPTIONAL -- the gate can
+#      shed events with its breaker closed and pass episode: nil -- so the
+#      warning silently vanished exactly when no episode existed.
+#   3. Written by upsert_storm_event, which runs AFTER the counts transaction
+#      commits and rescues its own failures. A transient save failure lost the
+#      marker while the batch ledger recorded the batch as applied, so the
+#      replay was suppressed and the gap was never recorded.
+#
+# So: its own table, written INSIDE the same transaction as the counts (either
+# both land or neither does), keyed by the interval it describes rather than by
+# an episode that may not exist.
+#
+# Bounded by construction: one row per (application, bucket) per degraded
+# flush, and only while the rollup is actually unusable -- which is a
+# misconfiguration, not a steady state. Retention prunes it by covered_until.
+class CreateEventTimingGaps < ActiveRecord::Migration[7.0]
+  def change
+    return if table_exists?(:rails_error_dashboard_event_timing_gaps)
+
+    create_table :rails_error_dashboard_event_timing_gaps do |t|
+      # Nullable: a gap can predate application scoping, and a NULL here means
+      # "applies to every application" rather than "unknown".
+      t.bigint :application_id
+      # The interval whose timing is unreliable. covered_from is the earliest
+      # event in the degraded flush, covered_until the latest.
+      t.datetime :covered_from, null: false
+      t.datetime :covered_until, null: false
+      # How many events lost their timestamps, so the UI can say how much of
+      # the window is affected rather than only that something is.
+      t.bigint :events_affected, null: false, default: 0
+      t.timestamps
+    end
+
+    # The read is "does any gap overlap the window I am displaying", which
+    # scans on covered_until. Named explicitly -- an auto-generated name here
+    # would exceed PostgreSQL's 63-character limit and fail during the HOST
+    # app's deploy (see mailboxer#480).
+    add_index :rails_error_dashboard_event_timing_gaps, :covered_until,
+              name: "index_red_timing_gaps_on_covered_until"
+  end
+end

data/lib/generators/rails_error_dashboard/install/templates/initializer.rb

@@ -21,6 +21,16 @@ RailsErrorDashboard.configure do |config|
   # Only notify (Slack, email, Discord, PagerDuty, webhooks, storm and
   # baseline alerts) for these environments. nil = every environment.
   # config.notification_environments = %w[production]
+  #
+  # Notification throttling. The per-error cooldown stops one error that keeps
+  # being reopened from paging repeatedly; it is held in the database, so it
+  # applies across every worker. The burst cap is for the deploy that throws
+  # hundreds of DIFFERENT new errors: after N new-error notifications in a
+  # window (per process) one summary message replaces the rest. Every error
+  # is still recorded.
+  # config.notification_cooldown_minutes = 5          # 0 = no cooldown
+  # config.notification_burst_limit = 10              # 0 = no cap
+  # config.notification_burst_window_seconds = 60
 
   # === Custom Authentication (optional) ===
   # Use your app's existing auth instead of HTTP Basic Auth.
@@ -61,7 +71,8 @@ RailsErrorDashboard.configure do |config|
   # User model for error associations
   config.user_model = "User"
 
-  # Error retention policy (days to keep errors before automatic deletion)
+  # Error retention policy: an error is deleted once it has not been seen for
+  # this many days (an error that is still occurring is never deleted)
   # Set to nil to keep errors forever (not recommended for production)
   # Run cleanup manually: rails error_dashboard:retention_cleanup
   # Or schedule the job: RailsErrorDashboard::RetentionCleanupJob.perform_later

data/lib/rails_error_dashboard/commands/assign_error.rb

@@ -4,7 +4,11 @@ module RailsErrorDashboard
   module Commands
     # Command: Assign an error to a user
     # This is a write operation that updates assignment fields on an ErrorLog record
+    # Returns {success: bool, error: ErrorLog}; a failure also carries
+    # reason: :blank_assignee and writes nothing.
     class AssignError
+      MAX_ASSIGNEE_LENGTH = 255
+
       def self.call(error_id, assigned_to:)
         new(error_id, assigned_to).call
       end
@@ -16,12 +20,18 @@ module RailsErrorDashboard
 
       def call
         error = ErrorLog.find(@error_id)
+
+        # A blank name used to "assign" the error to nobody and still move it to
+        # in_progress. A nested parameter is not a name either.
+        assignee = @assigned_to.is_a?(String) ? @assigned_to.strip.presence : nil
+        return { success: false, error: error, reason: :blank_assignee } unless assignee
+
         error.update!(
-          assigned_to: @assigned_to,
+          assigned_to: assignee.truncate(MAX_ASSIGNEE_LENGTH, omission: ""),
           assigned_at: Time.current,
           status: "in_progress" # Auto-transition to in_progress when assigned
         )
-        error
+        { success: true, error: error }
       end
     end
   end

data/lib/rails_error_dashboard/commands/backfill_environments.rb

@@ -39,6 +39,8 @@ module RailsErrorDashboard
             updated += ErrorLog.where(id: id, environment: nil).update_all(environment: env)
           end
         end
+        # Analytics groups by environment, and its result is cached.
+        Services::AnalyticsCacheManager.clear if updated.positive?
         updated
       end
 

data/lib/rails_error_dashboard/commands/backfill_resolved_at.rb

@@ -0,0 +1,42 @@
+# frozen_string_literal: true
+
+module RailsErrorDashboard
+  module Commands
+    # Command: give resolved errors that have no resolved_at a best-known one
+    #
+    # Errors resolved through the status workflow before 0.13.0 were marked
+    # resolved without a resolved_at, so MTTR never counted them. The moment of
+    # resolution was not recorded anywhere, but it was the row's last write
+    # unless something touched it afterwards, so updated_at is the closest
+    # available value.
+    #
+    # Behind `rake error_dashboard:backfill_resolved_at`. Idempotent: it only
+    # selects rows whose resolved_at is still NULL.
+    #
+    # @example
+    #   BackfillResolvedAt.call # => { updated: 42 }
+    class BackfillResolvedAt
+      BATCH_SIZE = 1000
+
+      def self.call(batch_size: BATCH_SIZE)
+        new(batch_size: batch_size).call
+      end
+
+      def initialize(batch_size: BATCH_SIZE)
+        @batch_size = batch_size
+      end
+
+      def call
+        updated = 0
+
+        ErrorLog.where(resolved: true, resolved_at: nil).in_batches(of: @batch_size) do |batch|
+          # SQL column reference, not a Ruby value: each row gets its own
+          # updated_at. update_all leaves updated_at itself alone.
+          updated += batch.update_all("resolved_at = updated_at")
+        end
+
+        { updated: updated }
+      end
+    end
+  end
+end

data/lib/rails_error_dashboard/commands/batch_delete_errors.rb

@@ -23,6 +23,7 @@ module RailsErrorDashboard
         error_ids_to_delete = errors.pluck(:id)
 
         errors.destroy_all
+        Services::AnalyticsCacheManager.clear
 
         # Dispatch plugin event for batch deleted errors
         PluginRegistry.dispatch(:on_errors_batch_deleted, error_ids_to_delete) if error_ids_to_delete.any?

data/lib/rails_error_dashboard/commands/batch_mute_errors.rb

@@ -41,6 +41,8 @@ module RailsErrorDashboard
           end
         end
 
+        Services::AnalyticsCacheManager.clear if muted_count.positive?
+
         PluginRegistry.dispatch(:on_errors_batch_muted, muted_errors) if muted_errors.any?
 
         {

data/lib/rails_error_dashboard/commands/batch_resolve_errors.rb

@@ -45,6 +45,8 @@ module RailsErrorDashboard
         end
 
         # Dispatch plugin event for batch resolved errors
+        Services::AnalyticsCacheManager.clear if resolved_count.positive?
+
         PluginRegistry.dispatch(:on_errors_batch_resolved, resolved_errors) if resolved_errors.any?
 
         {

data/lib/rails_error_dashboard/commands/batch_unmute_errors.rb

@@ -39,6 +39,8 @@ module RailsErrorDashboard
           end
         end
 
+        Services::AnalyticsCacheManager.clear if unmuted_count.positive?
+
         PluginRegistry.dispatch(:on_errors_batch_unmuted, unmuted_errors) if unmuted_errors.any?
 
         {