@oneuptime/common 12.0.33 → 12.0.34

package/build/dist/Types/Monitor/SeriesContext/SeriesLabelDisplay.js

@@ -0,0 +1,441 @@
+/*
+ * Turning a metric series' raw label map into something an on-call
+ * engineer can read at 3am.
+ *
+ * A grouped metric monitor stores the breaching series' identity on the
+ * alert/incident as `seriesLabels` - a map of ClickHouse attribute key
+ * to value, e.g.
+ *
+ *   { "resource.k8s.namespace.name": "prod",
+ *     "resource.k8s.pod.name": "checkout-7d9f-2xk",
+ *     "resource.k8s.container.name": "app",
+ *     "resource.k8s.node.name": "ip-10-0-3-14" }
+ *
+ * Those keys are correct but unreadable, and - more importantly - they
+ * used to live ONLY in a table on the alert detail page. The alert
+ * title, which is what shows up in the alert list, in Slack, in email
+ * and on a phone at 3am, carried none of it, so fifty alerts from one
+ * monitor rendered as fifty identical lines of text.
+ *
+ * This module is the shared, presentation-layer answer to that: given a
+ * label map it produces a stable, deduplicated, human-ordered rendering
+ * of the resource identity, in three shapes (inline summary, title
+ * suffix, markdown block). It lives in Types (not Server) because the
+ * dashboard renders the same identity next to alerts and must not
+ * reimplement the naming.
+ */
+/*
+ * Friendly names for the label keys OneUptime's own agents and the
+ * shipped recommendation templates actually group by.
+ *
+ * Keys are matched twice: once verbatim, and once with the ClickHouse
+ * `resource.` prefix stripped - OtelMetricsIngestService stamps that
+ * prefix onto OTel *resource* attributes, so the very same concept
+ * arrives as `resource.k8s.pod.name` from a resource attribute and as
+ * `k8s.pod.name` from a datapoint label. Registering the unprefixed
+ * spelling once covers both.
+ */
+const FriendlyLabelNames = {
+    // --- Kubernetes ---
+    "k8s.cluster.name": "Cluster",
+    "k8s.namespace.name": "Namespace",
+    "k8s.pod.name": "Pod",
+    "k8s.pod.uid": "Pod UID",
+    "k8s.container.name": "Container",
+    "k8s.node.name": "Node",
+    "k8s.deployment.name": "Deployment",
+    "k8s.replicaset.name": "ReplicaSet",
+    "k8s.statefulset.name": "StatefulSet",
+    "k8s.daemonset.name": "DaemonSet",
+    "k8s.job.name": "Job",
+    "k8s.cronjob.name": "CronJob",
+    "k8s.hpa.name": "HPA",
+    "k8s.persistentvolumeclaim.name": "PVC",
+    "k8s.volume.name": "Volume",
+    // --- Containers (Docker / Podman / Swarm) ---
+    "container.name": "Container",
+    "container.id": "Container ID",
+    "container.image.name": "Image",
+    "container.runtime": "Runtime",
+    "docker.swarm.service.name": "Swarm Service",
+    "docker.swarm.node.name": "Swarm Node",
+    "docker.swarm.cluster.name": "Swarm Cluster",
+    "oneuptime.docker.host.name": "Docker Host",
+    "oneuptime.podman.host.name": "Podman Host",
+    // --- Hosts / OS ---
+    "host.name": "Host",
+    "oneuptime.host.name": "Host",
+    "host.id": "Host ID",
+    mountpoint: "Mount",
+    device: "Device",
+    "system.device": "Device",
+    "disk.path": "Disk",
+    "network.interface": "Interface",
+    interfaceName: "Interface",
+    interfaceAlias: "Interface Alias",
+    diskPath: "Disk",
+    cpu: "CPU",
+    direction: "Direction",
+    state: "State",
+    process: "Process",
+    /*
+     * --- Proxmox ---
+     *
+     * pve-exporter encodes identity in a single `id` datapoint label
+     * ("qemu/100", "node/pve1", "storage/local"), and the agent's transform
+     * splits it into `pve.scope` / `pve.type` / `pve.id` (see
+     * ProxmoxAgent/otel-collector-config.yaml). Both spellings are named
+     * here because templates group by the raw `id` while a user's own
+     * monitor may group by the derived parts.
+     */
+    "proxmox.cluster.name": "Proxmox Cluster",
+    "proxmox.node.name": "Proxmox Node",
+    "proxmox.vm.name": "VM",
+    id: "Object",
+    "pve.id": "Object ID",
+    "pve.type": "Object Type",
+    "pve.scope": "Scope",
+    vmid: "VM ID",
+    node: "Node",
+    storage: "Storage",
+    // --- Ceph ---
+    "ceph.cluster.name": "Ceph Cluster",
+    ceph_daemon: "Ceph Daemon",
+    pool_id: "Pool",
+    pool: "Pool",
+    osd: "OSD",
+    // --- IoT ---
+    "iot.fleet.name": "Fleet",
+    "device.id": "Device",
+    "device.name": "Device",
+    // --- Services / RUM / telemetry ---
+    "service.name": "Service",
+    "oneuptime.service.name": "Service",
+    /*
+     * One process/task/replica of a service. What the ServiceAlertTemplates
+     * per-instance metric templates group by, so a "Thread Count Explosion"
+     * alert on a 40-replica service can say which replica. (The two request
+     * latency percentiles are deliberately ungrouped — they are service-level.)
+     */
+    "service.instance.id": "Instance",
+    "service.namespace": "Service Namespace",
+    "service.version": "Version",
+    "deployment.environment": "Environment",
+    "telemetry.sdk.language": "Language",
+    "url.path": "Path",
+    "http.route": "Route",
+    "browser.name": "Browser",
+    "os.name": "OS",
+    region: "Region",
+    "cloud.region": "Region",
+    "cloud.availability_zone": "Zone",
+};
+/*
+ * Ordering. An alert title has room for a couple of identifiers, and
+ * which couple it gets decides whether the title is useful: "Pod:
+ * checkout-7d9f-2xk" answers "what broke", "Namespace: prod" on its own
+ * does not. Lower number = more identifying = shown first and kept when
+ * the title is truncated.
+ *
+ * Container beats Pod deliberately for container-scoped metrics: a CPU
+ * *limit* is a property of one container, so when both labels are
+ * present the container is the thing that breached. Scope-setting
+ * labels (namespace, cluster, node) rank after the object itself - they
+ * qualify the answer rather than being it.
+ */
+const LabelPriority = {
+    "k8s.container.name": 10,
+    "container.name": 10,
+    "k8s.pod.name": 20,
+    "k8s.deployment.name": 25,
+    "k8s.statefulset.name": 25,
+    "k8s.daemonset.name": 25,
+    "k8s.job.name": 25,
+    "k8s.cronjob.name": 25,
+    "k8s.hpa.name": 25,
+    "k8s.persistentvolumeclaim.name": 25,
+    "k8s.volume.name": 25,
+    "docker.swarm.service.name": 25,
+    ceph_daemon: 25,
+    pool_id: 25,
+    "device.id": 25,
+    "device.name": 25,
+    "proxmox.vm.name": 25,
+    vmid: 25,
+    id: 30,
+    "pve.id": 32,
+    "pve.type": 55,
+    "pve.scope": 90,
+    mountpoint: 30,
+    diskPath: 30,
+    "disk.path": 30,
+    interfaceName: 30,
+    device: 35,
+    "system.device": 35,
+    storage: 35,
+    /*
+     * Above `service.name`: the service is already named by the monitor, the
+     * instance is the part the reader does not know.
+     */
+    "service.instance.id": 22,
+    "service.name": 40,
+    "oneuptime.service.name": 40,
+    "k8s.namespace.name": 50,
+    "service.namespace": 50,
+    "k8s.node.name": 60,
+    "proxmox.node.name": 60,
+    "docker.swarm.node.name": 60,
+    node: 60,
+    "host.name": 65,
+    "oneuptime.host.name": 65,
+    "oneuptime.docker.host.name": 65,
+    "oneuptime.podman.host.name": 65,
+    "k8s.cluster.name": 80,
+    "docker.swarm.cluster.name": 80,
+    "proxmox.cluster.name": 80,
+    "ceph.cluster.name": 80,
+    "iot.fleet.name": 80,
+    "k8s.pod.uid": 95,
+    "container.id": 95,
+};
+const DefaultLabelPriority = 70;
+/*
+ * How many identifiers a title carries before it stops being scannable.
+ * Three is enough for the two shapes that matter in practice -
+ * container/pod/namespace and pod/namespace/node - and the full set is
+ * always one click away in the description block and the alert's
+ * Affected Resource table.
+ */
+export const MaxLabelsInTitle = 3;
+/*
+ * Longest label value the inline (title) rendering will print in full.
+ *
+ * Kubernetes object names go up to 253 characters, and a generated pod
+ * name in a deeply-nested naming scheme really can get long. Three of
+ * those would overflow the 500-character `title` column and fail the
+ * alert INSERT outright - the alert would not exist at all, which is a
+ * far worse outcome than a shortened name. The full value is always
+ * present untruncated in the markdown block and in the alert's
+ * seriesLabels.
+ */
+export const MaxLabelValueLengthInTitle = 64;
+const Ellipsis = "...";
+export default class SeriesLabelDisplay {
+    /*
+     * Strip the ClickHouse `resource.` prefix so a resource attribute and
+     * the same attribute arriving as a datapoint label resolve to one
+     * entry in the tables above.
+     */
+    static normalizeKey(key) {
+        if (key.startsWith("resource.")) {
+            return key.slice("resource.".length);
+        }
+        return key;
+    }
+    /*
+     * "Pod", "Container", "Node" - or, for a key nobody registered, a
+     * best-effort prettified form of the key itself. An unknown key is
+     * still far more useful spelled out than dropped: a user who groups
+     * by `tenant_id` should see "Tenant Id: acme", not silence.
+     */
+    static getFriendlyLabelName(key) {
+        const normalized = SeriesLabelDisplay.normalizeKey(key);
+        const known = FriendlyLabelNames[normalized] || FriendlyLabelNames[key];
+        if (known) {
+            return known;
+        }
+        return SeriesLabelDisplay.prettifyKey(normalized);
+    }
+    static prettifyKey(key) {
+        const words = key
+            .split(/[._\-/]+/)
+            .filter((part) => {
+            return part.length > 0;
+        })
+            .map((part) => {
+            return part.charAt(0).toUpperCase() + part.slice(1);
+        });
+        if (words.length === 0) {
+            return key;
+        }
+        return words.join(" ");
+    }
+    /*
+     * Whether this key has a deliberately-registered name, as opposed to
+     * falling through to the prettified-key fallback.
+     *
+     * The fallback is right for a user's own attribute - nobody can
+     * enumerate what `oneuptime.captureMetric()` might emit - but a key a
+     * SHIPPED template groups by should always be registered, and there
+     * is no way to tell the two apart from the returned string alone
+     * (`device` prettifies to "Device", which is also its registered
+     * name). The recommendation-catalog tests use this to hold that line.
+     */
+    static isKnownLabelKey(key) {
+        const normalized = SeriesLabelDisplay.normalizeKey(key);
+        return (FriendlyLabelNames[normalized] !== undefined ||
+            FriendlyLabelNames[key] !== undefined);
+    }
+    static getLabelPriority(key) {
+        var _a;
+        const normalized = SeriesLabelDisplay.normalizeKey(key);
+        const priority = (_a = LabelPriority[normalized]) !== null && _a !== void 0 ? _a : LabelPriority[key];
+        return priority === undefined ? DefaultLabelPriority : priority;
+    }
+    /*
+     * Coerce one label value to the string a human should read.
+     *
+     * Multi-valued labels (a series grouped on an attribute that carries
+     * an array) join with ", ". Objects are refused rather than
+     * JSON-dumped into an alert title.
+     */
+    static toDisplayValue(value) {
+        if (value === undefined || value === null) {
+            return "";
+        }
+        if (Array.isArray(value)) {
+            return value
+                .map((entry) => {
+                return SeriesLabelDisplay.toDisplayValue(entry);
+            })
+                .filter((entry) => {
+                return entry.length > 0;
+            })
+                .join(", ");
+        }
+        if (typeof value === "object") {
+            return "";
+        }
+        return `${value}`.trim();
+    }
+    /*
+     * The display-ready labels for one series, most identifying first.
+     *
+     * Empty values are dropped, not rendered blank. A grouped series
+     * always carries every group-by key - `MetricSeriesFingerprint`
+     * deliberately preserves a key whose attribute was missing as `""` so
+     * the fingerprint stays stable across evaluations - so without this
+     * filter a pod whose node attribute happened to be absent would title
+     * itself "Pod: web-1 - Node: ".
+     *
+     * Values that are duplicates of an already-included value are dropped
+     * too: several spellings of the same identity routinely ride along
+     * (`k8s.pod.name` and `resource.k8s.pod.name`, `host.name` and
+     * `oneuptime.host.name`), and repeating them buys nothing.
+     */
+    static getDisplayLabels(seriesLabels) {
+        if (!seriesLabels) {
+            return [];
+        }
+        const labels = [];
+        const seenNameValuePairs = new Set();
+        const keys = Object.keys(seriesLabels);
+        // Stable order: priority first, then key, so equal-priority keys never flip.
+        keys.sort((a, b) => {
+            const priorityDifference = SeriesLabelDisplay.getLabelPriority(a) -
+                SeriesLabelDisplay.getLabelPriority(b);
+            if (priorityDifference !== 0) {
+                return priorityDifference;
+            }
+            return a.localeCompare(b);
+        });
+        for (const key of keys) {
+            const value = SeriesLabelDisplay.toDisplayValue(seriesLabels[key]);
+            if (!value) {
+                continue;
+            }
+            const name = SeriesLabelDisplay.getFriendlyLabelName(key);
+            const dedupeKey = `${name}${value}`;
+            if (seenNameValuePairs.has(dedupeKey)) {
+                continue;
+            }
+            seenNameValuePairs.add(dedupeKey);
+            labels.push({ key, name, value });
+        }
+        return labels;
+    }
+    /*
+     * "Pod: checkout-7d9f-2xk / Namespace: prod / Node: ip-10-0-3-14".
+     *
+     * Returns "" when the series carries no usable identity, so callers
+     * can concatenate unconditionally.
+     */
+    static buildInlineSummary(seriesLabels, options) {
+        var _a, _b;
+        const maxLabels = (_a = options === null || options === void 0 ? void 0 : options.maxLabels) !== null && _a !== void 0 ? _a : MaxLabelsInTitle;
+        const maxValueLength = (_b = options === null || options === void 0 ? void 0 : options.maxValueLength) !== null && _b !== void 0 ? _b : MaxLabelValueLengthInTitle;
+        const labels = SeriesLabelDisplay.getDisplayLabels(seriesLabels);
+        if (labels.length === 0) {
+            return "";
+        }
+        return labels
+            .slice(0, Math.max(maxLabels, 1))
+            .map((label) => {
+            return `${label.name}: ${SeriesLabelDisplay.truncateValue(label.value, maxValueLength)}`;
+        })
+            .join(" | ");
+    }
+    static truncateValue(value, maxLength) {
+        if (maxLength <= 0 || value.length <= maxLength) {
+            return value;
+        }
+        /*
+         * Keep the END of the value, not the start. Generated names put the
+         * distinguishing part last (`checkout-7d9f-2xk`), so a head-truncated
+         * `checkout-checkout-web-fro...` tells the reader nothing that the
+         * rest of the alert did not already.
+         */
+        if (maxLength <= Ellipsis.length) {
+            return value.slice(value.length - maxLength);
+        }
+        return `${Ellipsis}${value.slice(value.length - (maxLength - Ellipsis.length))}`;
+    }
+    /*
+     * The same summary, prefixed with the separator used in titles, or ""
+     * when there is nothing to say. Templates append this unconditionally:
+     * `"[K8s] Pod CPU Saturating Container Limit{{seriesResourceSuffix}}"`.
+     */
+    static buildTitleSuffix(seriesLabels, options) {
+        const summary = SeriesLabelDisplay.buildInlineSummary(seriesLabels, options);
+        if (!summary) {
+            return "";
+        }
+        return ` - ${summary}`;
+    }
+    /*
+     * A markdown block naming every identifier the series carries, for
+     * alert/incident descriptions (which is what Slack, email and the
+     * mobile push actually show).
+     */
+    static buildMarkdownBlock(seriesLabels, options) {
+        const labels = SeriesLabelDisplay.getDisplayLabels(seriesLabels);
+        if (labels.length === 0) {
+            return "";
+        }
+        const heading = (options === null || options === void 0 ? void 0 : options.heading) || "Affected resource";
+        const lines = labels.map((label) => {
+            return `- **${label.name}:** \`${label.value}\``;
+        });
+        return `**${heading}**\n${lines.join("\n")}`;
+    }
+    /*
+     * Look up one logical identity across every spelling it can arrive
+     * under. Callers pass unprefixed keys ("k8s.pod.name"); both the bare
+     * and the `resource.`-prefixed spelling are checked.
+     */
+    static findLabelValue(seriesLabels, keys) {
+        if (!seriesLabels) {
+            return "";
+        }
+        for (const key of keys) {
+            for (const candidate of [key, `resource.${key}`]) {
+                const value = SeriesLabelDisplay.toDisplayValue(seriesLabels[candidate]);
+                if (value) {
+                    return value;
+                }
+            }
+        }
+        return "";
+    }
+}
+//# sourceMappingURL=SeriesLabelDisplay.js.map

package/build/dist/Types/Monitor/SeriesContext/SeriesLabelDisplay.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"SeriesLabelDisplay.js","sourceRoot":"","sources":["../../../../../Types/Monitor/SeriesContext/SeriesLabelDisplay.ts"],"names":[],"mappings":"AAEA;;;;;;;;;;;;;;;;;;;;;;;;;GAyBG;AAEH;;;;;;;;;;GAUG;AACH,MAAM,kBAAkB,GAA2B;IACjD,qBAAqB;IACrB,kBAAkB,EAAE,SAAS;IAC7B,oBAAoB,EAAE,WAAW;IACjC,cAAc,EAAE,KAAK;IACrB,aAAa,EAAE,SAAS;IACxB,oBAAoB,EAAE,WAAW;IACjC,eAAe,EAAE,MAAM;IACvB,qBAAqB,EAAE,YAAY;IACnC,qBAAqB,EAAE,YAAY;IACnC,sBAAsB,EAAE,aAAa;IACrC,oBAAoB,EAAE,WAAW;IACjC,cAAc,EAAE,KAAK;IACrB,kBAAkB,EAAE,SAAS;IAC7B,cAAc,EAAE,KAAK;IACrB,gCAAgC,EAAE,KAAK;IACvC,iBAAiB,EAAE,QAAQ;IAE3B,+CAA+C;IAC/C,gBAAgB,EAAE,WAAW;IAC7B,cAAc,EAAE,cAAc;IAC9B,sBAAsB,EAAE,OAAO;IAC/B,mBAAmB,EAAE,SAAS;IAC9B,2BAA2B,EAAE,eAAe;IAC5C,wBAAwB,EAAE,YAAY;IACtC,2BAA2B,EAAE,eAAe;IAC5C,4BAA4B,EAAE,aAAa;IAC3C,4BAA4B,EAAE,aAAa;IAE3C,qBAAqB;IACrB,WAAW,EAAE,MAAM;IACnB,qBAAqB,EAAE,MAAM;IAC7B,SAAS,EAAE,SAAS;IACpB,UAAU,EAAE,OAAO;IACnB,MAAM,EAAE,QAAQ;IAChB,eAAe,EAAE,QAAQ;IACzB,WAAW,EAAE,MAAM;IACnB,mBAAmB,EAAE,WAAW;IAChC,aAAa,EAAE,WAAW;IAC1B,cAAc,EAAE,iBAAiB;IACjC,QAAQ,EAAE,MAAM;IAChB,GAAG,EAAE,KAAK;IACV,SAAS,EAAE,WAAW;IACtB,KAAK,EAAE,OAAO;IACd,OAAO,EAAE,SAAS;IAElB;;;;;;;;;OASG;IACH,sBAAsB,EAAE,iBAAiB;IACzC,mBAAmB,EAAE,cAAc;IACnC,iBAAiB,EAAE,IAAI;IACvB,EAAE,EAAE,QAAQ;IACZ,QAAQ,EAAE,WAAW;IACrB,UAAU,EAAE,aAAa;IACzB,WAAW,EAAE,OAAO;IACpB,IAAI,EAAE,OAAO;IACb,IAAI,EAAE,MAAM;IACZ,OAAO,EAAE,SAAS;IAElB,eAAe;IACf,mBAAmB,EAAE,cAAc;IACnC,WAAW,EAAE,aAAa;IAC1B,OAAO,EAAE,MAAM;IACf,IAAI,EAAE,MAAM;IACZ,GAAG,EAAE,KAAK;IAEV,cAAc;IACd,gBAAgB,EAAE,OAAO;IACzB,WAAW,EAAE,QAAQ;IACrB,aAAa,EAAE,QAAQ;IAEvB,qCAAqC;IACrC,cAAc,EAAE,SAAS;IACzB,wBAAwB,EAAE,SAAS;IACnC;;;;;OAKG;IACH,qBAAqB,EAAE,UAAU;IACjC,mBAAmB,EAAE,mBAAmB;IACxC,iBAAiB,EAAE,SAAS;IAC5B,wBAAwB,EAAE,aAAa;IACvC,wBAAwB,EAAE,UAAU;IACpC,UAAU,EAAE,MAAM;IAClB,YAAY,EAAE,OAAO;IACrB,cAAc,EAAE,SAAS;IACzB,SAAS,EAAE,IAAI;IACf,MAAM,EAAE,QAAQ;IAChB,cAAc,EAAE,QAAQ;IACxB,yBAAyB,EAAE,MAAM;CAClC,CAAC;AAEF;;;;;;;;;;;;GAYG;AACH,MAAM,aAAa,GAA2B;IAC5C,oBAAoB,EAAE,EAAE;IACxB,gBAAgB,EAAE,EAAE;IACpB,cAAc,EAAE,EAAE;IAClB,qBAAqB,EAAE,EAAE;IACzB,sBAAsB,EAAE,EAAE;IAC1B,oBAAoB,EAAE,EAAE;IACxB,cAAc,EAAE,EAAE;IAClB,kBAAkB,EAAE,EAAE;IACtB,cAAc,EAAE,EAAE;IAClB,gCAAgC,EAAE,EAAE;IACpC,iBAAiB,EAAE,EAAE;IACrB,2BAA2B,EAAE,EAAE;IAC/B,WAAW,EAAE,EAAE;IACf,OAAO,EAAE,EAAE;IACX,WAAW,EAAE,EAAE;IACf,aAAa,EAAE,EAAE;IACjB,iBAAiB,EAAE,EAAE;IACrB,IAAI,EAAE,EAAE;IACR,EAAE,EAAE,EAAE;IACN,QAAQ,EAAE,EAAE;IACZ,UAAU,EAAE,EAAE;IACd,WAAW,EAAE,EAAE;IACf,UAAU,EAAE,EAAE;IACd,QAAQ,EAAE,EAAE;IACZ,WAAW,EAAE,EAAE;IACf,aAAa,EAAE,EAAE;IACjB,MAAM,EAAE,EAAE;IACV,eAAe,EAAE,EAAE;IACnB,OAAO,EAAE,EAAE;IACX;;;OAGG;IACH,qBAAqB,EAAE,EAAE;IACzB,cAAc,EAAE,EAAE;IAClB,wBAAwB,EAAE,EAAE;IAC5B,oBAAoB,EAAE,EAAE;IACxB,mBAAmB,EAAE,EAAE;IACvB,eAAe,EAAE,EAAE;IACnB,mBAAmB,EAAE,EAAE;IACvB,wBAAwB,EAAE,EAAE;IAC5B,IAAI,EAAE,EAAE;IACR,WAAW,EAAE,EAAE;IACf,qBAAqB,EAAE,EAAE;IACzB,4BAA4B,EAAE,EAAE;IAChC,4BAA4B,EAAE,EAAE;IAChC,kBAAkB,EAAE,EAAE;IACtB,2BAA2B,EAAE,EAAE;IAC/B,sBAAsB,EAAE,EAAE;IAC1B,mBAAmB,EAAE,EAAE;IACvB,gBAAgB,EAAE,EAAE;IACpB,aAAa,EAAE,EAAE;IACjB,cAAc,EAAE,EAAE;CACnB,CAAC;AAEF,MAAM,oBAAoB,GAAW,EAAE,CAAC;AAExC;;;;;;GAMG;AACH,MAAM,CAAC,MAAM,gBAAgB,GAAW,CAAC,CAAC;AAE1C;;;;;;;;;;GAUG;AACH,MAAM,CAAC,MAAM,0BAA0B,GAAW,EAAE,CAAC;AAErD,MAAM,QAAQ,GAAW,KAAK,CAAC;AAW/B,MAAM,CAAC,OAAO,OAAO,kBAAkB;IACrC;;;;OAIG;IACI,MAAM,CAAC,YAAY,CAAC,GAAW;QACpC,IAAI,GAAG,CAAC,UAAU,CAAC,WAAW,CAAC,EAAE,CAAC;YAChC,OAAO,GAAG,CAAC,KAAK,CAAC,WAAW,CAAC,MAAM,CAAC,CAAC;QACvC,CAAC;QAED,OAAO,GAAG,CAAC;IACb,CAAC;IAED;;;;;OAKG;IACI,MAAM,CAAC,oBAAoB,CAAC,GAAW;QAC5C,MAAM,UAAU,GAAW,kBAAkB,CAAC,YAAY,CAAC,GAAG,CAAC,CAAC;QAEhE,MAAM,KAAK,GACT,kBAAkB,CAAC,UAAU,CAAC,IAAI,kBAAkB,CAAC,GAAG,CAAC,CAAC;QAE5D,IAAI,KAAK,EAAE,CAAC;YACV,OAAO,KAAK,CAAC;QACf,CAAC;QAED,OAAO,kBAAkB,CAAC,WAAW,CAAC,UAAU,CAAC,CAAC;IACpD,CAAC;IAEO,MAAM,CAAC,WAAW,CAAC,GAAW;QACpC,MAAM,KAAK,GAAkB,GAAG;aAC7B,KAAK,CAAC,UAAU,CAAC;aACjB,MAAM,CAAC,CAAC,IAAY,EAAE,EAAE;YACvB,OAAO,IAAI,CAAC,MAAM,GAAG,CAAC,CAAC;QACzB,CAAC,CAAC;aACD,GAAG,CAAC,CAAC,IAAY,EAAE,EAAE;YACpB,OAAO,IAAI,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,WAAW,EAAE,GAAG,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC;QACtD,CAAC,CAAC,CAAC;QAEL,IAAI,KAAK,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;YACvB,OAAO,GAAG,CAAC;QACb,CAAC;QAED,OAAO,KAAK,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;IACzB,CAAC;IAED;;;;;;;;;;OAUG;IACI,MAAM,CAAC,eAAe,CAAC,GAAW;QACvC,MAAM,UAAU,GAAW,kBAAkB,CAAC,YAAY,CAAC,GAAG,CAAC,CAAC;QAEhE,OAAO,CACL,kBAAkB,CAAC,UAAU,CAAC,KAAK,SAAS;YAC5C,kBAAkB,CAAC,GAAG,CAAC,KAAK,SAAS,CACtC,CAAC;IACJ,CAAC;IAEM,MAAM,CAAC,gBAAgB,CAAC,GAAW;;QACxC,MAAM,UAAU,GAAW,kBAAkB,CAAC,YAAY,CAAC,GAAG,CAAC,CAAC;QAEhE,MAAM,QAAQ,GACZ,MAAA,aAAa,CAAC,UAAU,CAAC,mCAAI,aAAa,CAAC,GAAG,CAAC,CAAC;QAElD,OAAO,QAAQ,KAAK,SAAS,CAAC,CAAC,CAAC,oBAAoB,CAAC,CAAC,CAAC,QAAQ,CAAC;IAClE,CAAC;IAED;;;;;;OAMG;IACK,MAAM,CAAC,cAAc,CAAC,KAA4B;QACxD,IAAI,KAAK,KAAK,SAAS,IAAI,KAAK,KAAK,IAAI,EAAE,CAAC;YAC1C,OAAO,EAAE,CAAC;QACZ,CAAC;QAED,IAAI,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC,EAAE,CAAC;YACzB,OAAO,KAAK;iBACT,GAAG,CAAC,CAAC,KAAgB,EAAE,EAAE;gBACxB,OAAO,kBAAkB,CAAC,cAAc,CAAC,KAAK,CAAC,CAAC;YAClD,CAAC,CAAC;iBACD,MAAM,CAAC,CAAC,KAAa,EAAE,EAAE;gBACxB,OAAO,KAAK,CAAC,MAAM,GAAG,CAAC,CAAC;YAC1B,CAAC,CAAC;iBACD,IAAI,CAAC,IAAI,CAAC,CAAC;QAChB,CAAC;QAED,IAAI,OAAO,KAAK,KAAK,QAAQ,EAAE,CAAC;YAC9B,OAAO,EAAE,CAAC;QACZ,CAAC;QAED,OAAO,GAAG,KAAK,EAAE,CAAC,IAAI,EAAE,CAAC;IAC3B,CAAC;IAED;;;;;;;;;;;;;;OAcG;IACI,MAAM,CAAC,gBAAgB,CAC5B,YAAoC;QAEpC,IAAI,CAAC,YAAY,EAAE,CAAC;YAClB,OAAO,EAAE,CAAC;QACZ,CAAC;QAED,MAAM,MAAM,GAA8B,EAAE,CAAC;QAC7C,MAAM,kBAAkB,GAAgB,IAAI,GAAG,EAAU,CAAC;QAE1D,MAAM,IAAI,GAAkB,MAAM,CAAC,IAAI,CAAC,YAAY,CAAC,CAAC;QAEtD,6EAA6E;QAC7E,IAAI,CAAC,IAAI,CAAC,CAAC,CAAS,EAAE,CAAS,EAAE,EAAE;YACjC,MAAM,kBAAkB,GACtB,kBAAkB,CAAC,gBAAgB,CAAC,CAAC,CAAC;gBACtC,kBAAkB,CAAC,gBAAgB,CAAC,CAAC,CAAC,CAAC;YAEzC,IAAI,kBAAkB,KAAK,CAAC,EAAE,CAAC;gBAC7B,OAAO,kBAAkB,CAAC;YAC5B,CAAC;YAED,OAAO,CAAC,CAAC,aAAa,CAAC,CAAC,CAAC,CAAC;QAC5B,CAAC,CAAC,CAAC;QAEH,KAAK,MAAM,GAAG,IAAI,IAAI,EAAE,CAAC;YACvB,MAAM,KAAK,GAAW,kBAAkB,CAAC,cAAc,CACrD,YAAY,CAAC,GAAG,CAAC,CAClB,CAAC;YAEF,IAAI,CAAC,KAAK,EAAE,CAAC;gBACX,SAAS;YACX,CAAC;YAED,MAAM,IAAI,GAAW,kBAAkB,CAAC,oBAAoB,CAAC,GAAG,CAAC,CAAC;YAElE,MAAM,SAAS,GAAW,GAAG,IAAI,IAAI,KAAK,EAAE,CAAC;YAE7C,IAAI,kBAAkB,CAAC,GAAG,CAAC,SAAS,CAAC,EAAE,CAAC;gBACtC,SAAS;YACX,CAAC;YAED,kBAAkB,CAAC,GAAG,CAAC,SAAS,CAAC,CAAC;YAElC,MAAM,CAAC,IAAI,CAAC,EAAE,GAAG,EAAE,IAAI,EAAE,KAAK,EAAE,CAAC,CAAC;QACpC,CAAC;QAED,OAAO,MAAM,CAAC;IAChB,CAAC;IAED;;;;;OAKG;IACI,MAAM,CAAC,kBAAkB,CAC9B,YAAoC,EACpC,OAKa;;QAEb,MAAM,SAAS,GAAW,MAAA,OAAO,aAAP,OAAO,uBAAP,OAAO,CAAE,SAAS,mCAAI,gBAAgB,CAAC;QACjE,MAAM,cAAc,GAClB,MAAA,OAAO,aAAP,OAAO,uBAAP,OAAO,CAAE,cAAc,mCAAI,0BAA0B,CAAC;QAExD,MAAM,MAAM,GACV,kBAAkB,CAAC,gBAAgB,CAAC,YAAY,CAAC,CAAC;QAEpD,IAAI,MAAM,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;YACxB,OAAO,EAAE,CAAC;QACZ,CAAC;QAED,OAAO,MAAM;aACV,KAAK,CAAC,CAAC,EAAE,IAAI,CAAC,GAAG,CAAC,SAAS,EAAE,CAAC,CAAC,CAAC;aAChC,GAAG,CAAC,CAAC,KAAyB,EAAE,EAAE;YACjC,OAAO,GAAG,KAAK,CAAC,IAAI,KAAK,kBAAkB,CAAC,aAAa,CACvD,KAAK,CAAC,KAAK,EACX,cAAc,CACf,EAAE,CAAC;QACN,CAAC,CAAC;aACD,IAAI,CAAC,KAAK,CAAC,CAAC;IACjB,CAAC;IAEO,MAAM,CAAC,aAAa,CAAC,KAAa,EAAE,SAAiB;QAC3D,IAAI,SAAS,IAAI,CAAC,IAAI,KAAK,CAAC,MAAM,IAAI,SAAS,EAAE,CAAC;YAChD,OAAO,KAAK,CAAC;QACf,CAAC;QAED;;;;;WAKG;QACH,IAAI,SAAS,IAAI,QAAQ,CAAC,MAAM,EAAE,CAAC;YACjC,OAAO,KAAK,CAAC,KAAK,CAAC,KAAK,CAAC,MAAM,GAAG,SAAS,CAAC,CAAC;QAC/C,CAAC;QAED,OAAO,GAAG,QAAQ,GAAG,KAAK,CAAC,KAAK,CAAC,KAAK,CAAC,MAAM,GAAG,CAAC,SAAS,GAAG,QAAQ,CAAC,MAAM,CAAC,CAAC,EAAE,CAAC;IACnF,CAAC;IAED;;;;OAIG;IACI,MAAM,CAAC,gBAAgB,CAC5B,YAAoC,EACpC,OAKa;QAEb,MAAM,OAAO,GAAW,kBAAkB,CAAC,kBAAkB,CAC3D,YAAY,EACZ,OAAO,CACR,CAAC;QAEF,IAAI,CAAC,OAAO,EAAE,CAAC;YACb,OAAO,EAAE,CAAC;QACZ,CAAC;QAED,OAAO,MAAM,OAAO,EAAE,CAAC;IACzB,CAAC;IAED;;;;OAIG;IACI,MAAM,CAAC,kBAAkB,CAC9B,YAAoC,EACpC,OAAsD;QAEtD,MAAM,MAAM,GACV,kBAAkB,CAAC,gBAAgB,CAAC,YAAY,CAAC,CAAC;QAEpD,IAAI,MAAM,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;YACxB,OAAO,EAAE,CAAC;QACZ,CAAC;QAED,MAAM,OAAO,GAAW,CAAA,OAAO,aAAP,OAAO,uBAAP,OAAO,CAAE,OAAO,KAAI,mBAAmB,CAAC;QAEhE,MAAM,KAAK,GAAkB,MAAM,CAAC,GAAG,CAAC,CAAC,KAAyB,EAAE,EAAE;YACpE,OAAO,OAAO,KAAK,CAAC,IAAI,SAAS,KAAK,CAAC,KAAK,IAAI,CAAC;QACnD,CAAC,CAAC,CAAC;QAEH,OAAO,KAAK,OAAO,OAAO,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC;IAC/C,CAAC;IAED;;;;OAIG;IACI,MAAM,CAAC,cAAc,CAC1B,YAAoC,EACpC,IAA2B;QAE3B,IAAI,CAAC,YAAY,EAAE,CAAC;YAClB,OAAO,EAAE,CAAC;QACZ,CAAC;QAED,KAAK,MAAM,GAAG,IAAI,IAAI,EAAE,CAAC;YACvB,KAAK,MAAM,SAAS,IAAI,CAAC,GAAG,EAAE,YAAY,GAAG,EAAE,CAAC,EAAE,CAAC;gBACjD,MAAM,KAAK,GAAW,kBAAkB,CAAC,cAAc,CACrD,YAAY,CAAC,SAAS,CAAC,CACxB,CAAC;gBAEF,IAAI,KAAK,EAAE,CAAC;oBACV,OAAO,KAAK,CAAC;gBACf,CAAC;YACH,CAAC;QACH,CAAC;QAED,OAAO,EAAE,CAAC;IACZ,CAAC;CACF"}

package/build/dist/Types/Monitor/ServiceAlertTemplates.js

@@ -9,16 +9,19 @@ import MonitorType from "./MonitorType";
 import RollingTime from "../RollingTime/RollingTime";
 import MetricsAggregationType from "../Metrics/MetricsAggregationType";
 /*
- * `EvaluateOverTimeType` names the window collapse, and its two members used
- * here do NOT mean what their names suggest — worth stating once, here, rather
- * than being rediscovered per template:
+ * `EvaluateOverTimeType` names the window collapse. The two members these
+ * templates use are worth stating once, here, rather than being rediscovered
+ * per template:
  *
  *   AnyValue  -> ANY bucket in the rolling window breaching is a breach.
  *                Right for spikes.
- *   AllValues -> EVERY bucket must breach. Right for sustained pressure, and
- *                what the comparators actually do for every member that is not
- *                `AnyValue` (they fall through to `.every()`; there is no
- *                averaging on this code path despite `Average` existing).
+ *   AllValues -> EVERY bucket must breach. Right for sustained pressure.
+ *
+ * The reducing members (`Average`, `Sum`, `MaximumValue`, `MunimumValue`) do
+ * now collapse the window to one sample before the comparison —
+ * `CompareCriteria.reduceWindow` — so they mean what they say. `AllValues` is
+ * still the right default here: a reduced window hides the shape, and a metric
+ * that spends half the window fine is not under sustained pressure.
  *
  * A window with no samples at all never reaches the comparator — the
  * evaluator's no-data guard returns "not breaching" under the default
@@ -32,6 +35,33 @@ import MetricsAggregationType from "../Metrics/MetricsAggregationType";
  * team to mute it.
  */
 const SUSTAINED = EvaluateOverTimeType.AllValues;
+/*
+ * The resource attribute that identifies ONE process of a service.
+ *
+ * Every metric template here except the two request-latency percentiles
+ * thresholds a property of a single process — a heap, an event loop, a
+ * goroutine count, a thread pool. Without a group-by the ClickHouse
+ * aggregation collapses every replica of the service into one number, so an
+ * `Avg` template on three replicas at 95/20/20 reads 45 and stays green while
+ * one of them OOM-kills, and a `Max` template fires on the worst replica with
+ * no way to say which. Grouping fixes both at once: the worker emits one
+ * series per instance and the alert carries that instance in `seriesLabels`,
+ * which `SeriesLabelDisplay` renders into the title and the description block.
+ *
+ * `service.instance.id` is OPTIONAL in the resource semantic conventions, and
+ * this degrades safely when it is absent: the query returns one series whose
+ * label value is the empty string (ClickHouse `attributes[k]` on a missing key
+ * yields ""), `MetricSeriesFingerprint.extractSeriesLabels` keeps the key so
+ * the fingerprint stays stable across evaluations, and
+ * `SeriesLabelDisplay.getDisplayLabels` drops empty values rather than
+ * rendering "Instance: ". A service that does not report it therefore gets
+ * exactly today's single-series behaviour, not silence.
+ *
+ * The `resource.` prefix is the one OtelMetricsIngestService stamps onto OTel
+ * resource attributes (`prefixKeysWithString: "resource"`), the same spelling
+ * KubernetesAlertTemplates uses for `resource.k8s.pod.name`.
+ */
+const PER_INSTANCE_GROUP_BY = ["resource.service.instance.id"];
 /*
  * The unhealthy/healthy criteria pair every template ships.
  *
@@ -58,7 +88,16 @@ function buildCriteriaPair(data) {
                 checkOn: data.checkOn,
                 filterType: data.unhealthyFilterType,
                 value: data.threshold,
-                metricMonitorOptions: metricMonitorOptions,
+                /*
+                 * A copy per filter, not the shared object. `MetricMonitorCriteria`
+                 * writes its `AnyValue` default straight back through this reference
+                 * when `metricAggregationType` is unset, and the criteria form spreads
+                 * and re-assigns it on edit — either of which, on a shared object,
+                 * changes the healthy criteria as a side effect of touching the
+                 * unhealthy one.
+                 */
+                metricMonitorOptions: metricMonitorOptions
+                    ? Object.assign({}, metricMonitorOptions) : undefined,
             },
         ],
         incidents: [
@@ -97,7 +136,9 @@ function buildCriteriaPair(data) {
                 checkOn: data.checkOn,
                 filterType: data.healthyFilterType,
                 value: data.threshold,
-                metricMonitorOptions: metricMonitorOptions,
+                // A copy per filter — see the unhealthy filter above.
+                metricMonitorOptions: metricMonitorOptions
+                    ? Object.assign({}, metricMonitorOptions) : undefined,
             },
         ],
         incidents: [],
@@ -168,14 +209,14 @@ function buildQueryConfig(data) {
             legend: data.title,
             legendUnit: data.legendUnit,
         },
-        metricQueryData: {
-            filterData: {
+        metricQueryData: Object.assign({ filterData: {
                 metricName: data.metricName,
                 attributes: data.attributes || {},
                 aggegationType: data.aggregationType,
                 aggregateBy: {},
-            },
-        },
+            } }, (data.groupByAttributeKeys && data.groupByAttributeKeys.length > 0
+            ? { groupByAttributeKeys: data.groupByAttributeKeys }
+            : {})),
     };
 }
 /*
@@ -211,6 +252,7 @@ function buildMetricTemplate(data) {
                             aggregationType: data.aggregationType,
                             attributes: data.attributes,
                             legendUnit: data.legendUnit,
+                            groupByAttributeKeys: data.groupByAttributeKeys || PER_INSTANCE_GROUP_BY,
                         }),
                     ],
                     formulaConfigs: [],
@@ -293,6 +335,17 @@ function buildRatioTemplate(data) {
                             title: `${data.name} (used)`,
                             aggregationType: MetricsAggregationType.Sum,
                             attributes: data.numeratorAttributes,
+                            /*
+                             * The SAME key set on both queries. The worker splits every
+                             * query on the UNION of their keys, so the split itself cannot
+                             * go wrong from setting only one side — but
+                             * `MetricMonitorCriteria.buildContext` reads
+                             * `groupByAttributeKeys` off a SINGLE query config when it
+                             * builds the alert's "Grouped By" root-cause block, and a
+                             * one-sided setting renders that block empty for whichever side
+                             * it read.
+                             */
+                            groupByAttributeKeys: data.groupByAttributeKeys || PER_INSTANCE_GROUP_BY,
                         }),
                         buildQueryConfig({
                             metricName: data.denominatorMetricName,
@@ -300,6 +353,7 @@ function buildRatioTemplate(data) {
                             title: `${data.name} (limit)`,
                             aggregationType: MetricsAggregationType.Sum,
                             attributes: data.denominatorAttributes,
+                            groupByAttributeKeys: data.groupByAttributeKeys || PER_INSTANCE_GROUP_BY,
                         }),
                     ],
                     formulaConfigs: [formulaConfig],
@@ -511,6 +565,14 @@ const latencyP95Template = buildMetricTemplate({
     metricAlias: "service_latency_p95",
     aggregationType: MetricsAggregationType.P95,
     legendUnit: "ms",
+    /*
+     * Not grouped, unlike every other metric template here. A request latency
+     * percentile is a property of the SERVICE — it is the number the SLO is
+     * written against — and splitting it per instance would turn "the service's
+     * p95 is above a second" into "one replica's p95 is above a second", which
+     * fires on a single slow pod and means something else.
+     */
+    groupByAttributeKeys: [],
     threshold: 1000,
     thresholdLabel: "1,000 ms",
     incidentDescription: "One in twenty requests is now taking a second or more. Break the latency down by route and by downstream call to find where the time is going.",
@@ -525,6 +587,8 @@ const latencyP99Template = buildMetricTemplate({
     metricAlias: "service_latency_p99",
     aggregationType: MetricsAggregationType.P99,
     legendUnit: "ms",
+    // Service-level, not per instance — see latencyP95Template.
+    groupByAttributeKeys: [],
     threshold: 2500,
     thresholdLabel: "2,500 ms",
     incidentDescription: "The slowest one percent of requests are taking 2.5 seconds or more. Look for a saturated dependency, a lock, or a slow query on the affected route.",
@@ -849,7 +913,7 @@ const pythonTemplates = [
     buildMetricTemplate({
         id: "service-python-rss-memory",
         name: "Resident Memory High",
-        description: "Alert when resident memory stays above 1 GB per worker — the signal that precedes an OOM kill. Needs the system-metrics instrumentation enabled.",
+        description: "Alert when a worker's resident memory stays above 1 GB — the signal that precedes an OOM kill. Needs the system-metrics instrumentation enabled. Reads process.runtime.cpython.memory with type=rss; newer opentelemetry-python releases report this as process.memory.usage, which the language-agnostic Process Memory High template already covers.",
         category: "Python Runtime",
         severity: "Warning",
         language: "python",
@@ -866,7 +930,7 @@ const pythonTemplates = [
     buildMetricTemplate({
         id: "service-python-thread-growth",
         name: "Thread Count Growth",
-        description: "Alert when thread count stays above 200, which usually means a leaked executor or an unclosed client pool. Needs the system-metrics instrumentation enabled.",
+        description: "Alert when thread count stays above 200, which usually means a leaked executor or an unclosed client pool. Needs the system-metrics instrumentation enabled. Reads process.runtime.cpython.thread_count; newer opentelemetry-python releases report this as process.thread.count — retarget the created monitor in one edit if your SDK emits that instead.",
         category: "Python Runtime",
         severity: "Warning",
         language: "python",