v1.9.42: full Grafana/Prometheus GUI & CLI improvements

Grafana tab:
- Datasources table (Name, Type, URL, Default)
- Open Grafana button (opens browser)
- Switch to AlertManager endpoint for real-time active alerts

Prometheus tab:
- Quick query buttons (up, CPU, Memory)
- Metrics browser popup with filter (loads all metric names)
- Rules section (recording + alerting rules Treeview)

CLI:
- --grafana-datasources ALIAS
- --prom-rules ALIAS

i18n: 28 new keys (EN/RU/ZH)

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
This commit is contained in:
chrome-storm-c442
2026-03-06 10:23:33 -05:00
parent 6a2db542c3
commit 8ff62d8f11
7 changed files with 384 additions and 61 deletions

View File

@@ -1,5 +1,5 @@
"""
Prometheus tab — PromQL query executor, targets overview, and alerts.
Prometheus tab — PromQL query executor, targets overview, alerts, and rules.
"""
import threading
@@ -8,7 +8,7 @@ from tkinter import ttk
import customtkinter as ctk
from core.prometheus_client import PrometheusClient
from core.i18n import t
from core.icons import icon_text, make_icon_button, reconfigure_icon_button
from core.icons import make_icon_button, reconfigure_icon_button
from gui.tabs.query_tab import apply_dark_scrollbar_style
@@ -41,19 +41,35 @@ class PrometheusTab(ctk.CTkFrame):
command=self._execute_query)
self._exec_btn.pack(side="left")
# ── Quick query buttons ──
quick_frame = ctk.CTkFrame(self, fg_color="transparent")
quick_frame.pack(fill="x", padx=15, pady=(0, 5))
for label, query in [("up", "up"), ("CPU", "process_cpu_seconds_total"),
("Memory", "node_memory_MemFree_bytes")]:
btn = make_icon_button(quick_frame, "metrics", label, width=80,
fg_color="#6b7280", hover_color="#4b5563",
command=lambda q=query: self._run_quick(q))
btn.pack(side="left", padx=(0, 5))
self._metrics_btn = make_icon_button(quick_frame, "search", t("prom_metrics_browser"), width=100,
fg_color="#6b7280", hover_color="#4b5563",
command=self._open_metrics_browser)
self._metrics_btn.pack(side="left", padx=(0, 5))
# ── Query results ──
results_label = ctk.CTkLabel(self, text=t("prom_results"),
font=ctk.CTkFont(size=12, weight="bold"), anchor="w")
results_label.pack(fill="x", padx=15, pady=(10, 3))
results_label.pack(fill="x", padx=15, pady=(5, 3))
self._results_box = ctk.CTkTextbox(self, height=150,
self._results_box = ctk.CTkTextbox(self, height=120,
font=ctk.CTkFont(family="Consolas", size=12),
state="disabled")
self._results_box.pack(fill="x", padx=15, pady=(0, 5))
# ── Targets section ──
targets_header = ctk.CTkFrame(self, fg_color="transparent")
targets_header.pack(fill="x", padx=15, pady=(10, 3))
targets_header.pack(fill="x", padx=15, pady=(5, 3))
targets_label = ctk.CTkLabel(targets_header, text=t("prom_targets"),
font=ctk.CTkFont(size=14, weight="bold"), anchor="w")
@@ -68,7 +84,7 @@ class PrometheusTab(ctk.CTkFrame):
target_columns = ("job", "instance", "health", "last_scrape")
self._targets_tree = ttk.Treeview(targets_frame, columns=target_columns, show="headings",
selectmode="browse", height=6)
selectmode="browse", height=5)
self._targets_tree.heading("job", text=t("prom_target_job"))
self._targets_tree.heading("instance", text=t("prom_target_instance"))
self._targets_tree.heading("health", text=t("prom_target_health"))
@@ -88,13 +104,39 @@ class PrometheusTab(ctk.CTkFrame):
# ── Alerts section ──
alerts_label = ctk.CTkLabel(self, text=t("prom_alerts"),
font=ctk.CTkFont(size=14, weight="bold"), anchor="w")
alerts_label.pack(fill="x", padx=15, pady=(10, 3))
alerts_label.pack(fill="x", padx=15, pady=(5, 3))
self._alerts_box = ctk.CTkTextbox(self, height=100,
self._alerts_box = ctk.CTkTextbox(self, height=80,
font=ctk.CTkFont(family="Consolas", size=12),
state="disabled")
self._alerts_box.pack(fill="x", padx=15, pady=(0, 5))
# ── Rules section ──
rules_label = ctk.CTkLabel(self, text=t("prom_rules"),
font=ctk.CTkFont(size=14, weight="bold"), anchor="w")
rules_label.pack(fill="x", padx=15, pady=(5, 3))
rules_frame = ctk.CTkFrame(self, fg_color="transparent")
rules_frame.pack(fill="both", expand=True, padx=15, pady=(0, 5))
rules_columns = ("type", "name", "group", "health")
self._rules_tree = ttk.Treeview(rules_frame, columns=rules_columns, show="headings",
selectmode="browse", height=5)
self._rules_tree.heading("type", text=t("prom_rule_type"))
self._rules_tree.heading("name", text=t("prom_rule_name"))
self._rules_tree.heading("group", text=t("prom_rule_group"))
self._rules_tree.heading("health", text=t("prom_rule_health"))
self._rules_tree.column("type", width=80, minwidth=60)
self._rules_tree.column("name", width=250, minwidth=120)
self._rules_tree.column("group", width=150, minwidth=80)
self._rules_tree.column("health", width=80, minwidth=60)
self._rules_tree.pack(side="left", fill="both", expand=True)
rules_scroll = ttk.Scrollbar(rules_frame, orient="vertical", command=self._rules_tree.yview,
style="Dark.Vertical.TScrollbar")
rules_scroll.pack(side="right", fill="y")
self._rules_tree.configure(yscrollcommand=rules_scroll.set)
# ── Status bar ──
self._status_bar = ctk.CTkLabel(self, text=t("prom_no_server"), anchor="w",
font=ctk.CTkFont(size=11), text_color="#9ca3af")
@@ -103,7 +145,6 @@ class PrometheusTab(ctk.CTkFrame):
# ── Public API ──
def set_server(self, alias: str | None):
"""Called when user selects a server in sidebar."""
self._current_alias = alias
self._client = None
self._clear_all()
@@ -114,6 +155,13 @@ class PrometheusTab(ctk.CTkFrame):
else:
self._set_status(t("prom_no_server"), "#9ca3af")
# ── Quick query ──
def _run_quick(self, query: str):
self._query_entry.delete(0, "end")
self._query_entry.insert(0, query)
self._execute_query()
# ── PromQL execution ──
def _execute_query(self):
@@ -121,7 +169,7 @@ class PrometheusTab(ctk.CTkFrame):
if not query:
return
if not self._current_alias:
self._set_results(t("no_server_selected"))
self._set_results(t("prom_no_server"))
return
self._exec_btn.configure(state="disabled")
@@ -141,11 +189,9 @@ class PrometheusTab(ctk.CTkFrame):
threading.Thread(target=_do, daemon=True).start()
def _format_query_result(self, result: dict) -> str:
"""Format Prometheus query API response for display."""
status = result.get("status", "unknown")
if status != "success":
error = result.get("error", "Unknown error")
return f"Error: {error}"
return f"Error: {result.get('error', 'Unknown error')}"
data = result.get("data", {})
result_type = data.get("resultType", "")
@@ -166,7 +212,7 @@ class PrometheusTab(ctk.CTkFrame):
elif result_type == "matrix":
values = item.get("values", [])
lines.append(f"{{{metric_str}}}")
for ts, val in values[-10:]: # Show last 10 points
for ts, val in values[-10:]:
lines.append(f" @{ts} => {val}")
if len(values) > 10:
lines.append(f" ... ({len(values)} total points)")
@@ -175,14 +221,84 @@ class PrometheusTab(ctk.CTkFrame):
return "\n".join(lines)
# ── Refresh targets & alerts ──
# ── Metrics browser ──
def _open_metrics_browser(self):
if not self._current_alias:
self._set_status(t("prom_no_server"), "#ef4444")
return
self._metrics_btn.configure(state="disabled")
def _do():
try:
client = self._get_client()
resp = client._get("/api/v1/label/__name__/values")
metrics = resp.get("data", [])
self.after(0, lambda: self._show_metrics_popup(metrics))
except Exception as e:
self.after(0, lambda: self._set_status(f"(error) {e}", "#ef4444"))
finally:
self.after(0, lambda: self._metrics_btn.configure(state="normal"))
threading.Thread(target=_do, daemon=True).start()
def _show_metrics_popup(self, metrics: list[str]):
popup = ctk.CTkToplevel(self)
popup.title(t("prom_metrics_browser"))
popup.geometry("450x500")
popup.transient(self.winfo_toplevel())
filter_entry = ctk.CTkEntry(popup, placeholder_text=t("prom_filter_metrics"))
filter_entry.pack(fill="x", padx=10, pady=(10, 5))
listbox_frame = ctk.CTkFrame(popup, fg_color="transparent")
listbox_frame.pack(fill="both", expand=True, padx=10, pady=(0, 10))
tree = ttk.Treeview(listbox_frame, columns=("metric",), show="headings",
selectmode="browse")
tree.heading("metric", text="Metric Name")
tree.column("metric", width=400)
tree.pack(side="left", fill="both", expand=True)
scroll = ttk.Scrollbar(listbox_frame, orient="vertical", command=tree.yview,
style="Dark.Vertical.TScrollbar")
scroll.pack(side="right", fill="y")
tree.configure(yscrollcommand=scroll.set)
all_metrics = sorted(metrics)
def populate(filter_text=""):
tree.delete(*tree.get_children())
for m in all_metrics:
if filter_text.lower() in m.lower():
tree.insert("", "end", values=(m,))
def on_filter(*_):
populate(filter_entry.get())
filter_entry.bind("<KeyRelease>", on_filter)
def on_select(event):
sel = tree.selection()
if sel:
metric = tree.item(sel[0])["values"][0]
self._query_entry.delete(0, "end")
self._query_entry.insert(0, metric)
popup.destroy()
tree.bind("<Double-1>", on_select)
populate()
filter_entry.focus_set()
# ── Refresh targets, alerts & rules ──
def _refresh_all(self):
if not self._current_alias:
self._set_status(t("no_server_selected"), "#ef4444")
self._set_status(t("prom_no_server"), "#ef4444")
return
self._refresh_btn.configure(state="disabled", text=t("prom_loading"))
self._refresh_btn.configure(state="disabled")
self._set_status(t("prom_loading"), "#ccaa00")
def _do():
@@ -193,12 +309,17 @@ class PrometheusTab(ctk.CTkFrame):
targets = targets_resp.get("data", {}).get("activeTargets", [])
alerts_resp = client.alerts()
alerts = alerts_resp.get("data", {}).get("alerts", [])
rules_resp = client.rules()
rule_groups = rules_resp.get("data", {}).get("groups", [])
self.after(0, lambda: self._populate_targets(targets))
self.after(0, lambda: self._populate_alerts(alerts))
self.after(0, lambda: self._populate_rules(rule_groups))
rule_count = sum(len(g.get("rules", [])) for g in rule_groups)
self.after(0, lambda: self._set_status(
t("prom_loaded").format(
targets=len(targets), alerts=len(alerts)
targets=len(targets), alerts=len(alerts), rules=rule_count
), "#22c55e"))
except Exception as e:
self.after(0, lambda: self._set_status(f"(error) {e}", "#ef4444"))
@@ -223,28 +344,19 @@ class PrometheusTab(ctk.CTkFrame):
def _populate_targets(self, targets: list[dict]):
self._targets_tree.delete(*self._targets_tree.get_children())
for target in targets:
job = target.get("labels", {}).get("job", "")
instance = target.get("labels", {}).get("instance", "")
job = target.get("labels", {}).get("job", "---")
instance = target.get("labels", {}).get("instance", "---")
health = target.get("health", "unknown")
last_scrape = target.get("lastScrape", "")
tag = ""
if health == "up":
tag = "up"
elif health == "down":
tag = "down"
last_scrape = target.get("lastScrape", "---")
tag = "up" if health == "up" else ("down" if health == "down" else "")
self._targets_tree.insert("", "end",
values=(job, instance, health, last_scrape),
tags=(tag,))
values=(job, instance, health, last_scrape), tags=(tag,))
self._targets_tree.tag_configure("up", foreground="#22c55e")
self._targets_tree.tag_configure("down", foreground="#ef4444")
def _populate_alerts(self, alerts: list[dict]):
self._alerts_box.configure(state="normal")
self._alerts_box.delete("1.0", "end")
if not alerts:
self._alerts_box.insert("1.0", t("prom_no_alerts"))
else:
@@ -252,12 +364,25 @@ class PrometheusTab(ctk.CTkFrame):
for a in alerts:
name = a.get("labels", {}).get("alertname", a.get("name", "unknown"))
state = a.get("state", "unknown")
severity = a.get("labels", {}).get("severity", "")
severity = a.get("labels", {}).get("severity", "---")
lines.append(f"[{state.upper()}] {name} (severity: {severity})")
self._alerts_box.insert("1.0", "\n".join(lines))
self._alerts_box.configure(state="disabled")
def _populate_rules(self, groups: list[dict]):
self._rules_tree.delete(*self._rules_tree.get_children())
for group in groups:
group_name = group.get("name", "")
for rule in group.get("rules", []):
rtype = rule.get("type", "")
name = rule.get("name", "")
health = rule.get("health", "")
tag = "up" if health == "ok" else ("down" if health == "err" else "")
self._rules_tree.insert("", "end",
values=(rtype, name, group_name, health), tags=(tag,))
self._rules_tree.tag_configure("up", foreground="#22c55e")
self._rules_tree.tag_configure("down", foreground="#ef4444")
# ── Helpers ──
def _set_results(self, text: str):
@@ -268,6 +393,7 @@ class PrometheusTab(ctk.CTkFrame):
def _clear_all(self):
self._targets_tree.delete(*self._targets_tree.get_children())
self._rules_tree.delete(*self._rules_tree.get_children())
self._set_results("")
self._alerts_box.configure(state="normal")
self._alerts_box.delete("1.0", "end")