2 h (S10.1 : 25 min · S10.2 : 30 min · S10.3 : 40 min · S10.4 : 25 min)
modules 3 (log-stNN-shared), 7 (web and API VMs, Azure Monitor Agent) and 9 (vault rsv-stNN-arveo)
monitoring prepared at 13:30 by preparer-supervision.sh (VMs started, diagnostic settings, flow log)
25 min
Monitoring = Collect + Analyze + Alert + ActBefore (Lyon): customer call → login to each server → cause found (hours)
After (Azure): alert in 5 min → KQL query on all servers → cause found (minutes)
Azure Monitor = Sources → Metrics | Logs → Alerts + Analysis + Visualizationflowchart LR R[Azure resources<br/>platform] --> M[(Metrics<br/>93 days)] R -->|diagnostic setting| L[(Logs<br/>Log Analytics)] V[VMs<br/>agent + DCR] --> L J[Activity log<br/>subscription] -.->|export| L M --> A[Alerts] L --> A L --> K[KQL, workbooks]
Metric = Timestamped numeric value · Log = Structured record in a table| Criterion | Metrics | Logs |
|---|---|---|
| Content | Numeric values at regular intervals | Events, text, varied properties |
| Latency | Near real time (about 1 min) | Minutes (ingestion) [TO VERIFY] delays |
| Retention | 93 days, free of charge | Configurable (30 days here), billed per GB |
| Analysis | Metrics Explorer, charts | KQL: filters, joins, grouping |
| Alerts | Evaluated every minute, low cost | Scheduled query, slower and costlier |
Sources = Platform + Resources + Activity + Guest OS + Application| Source | Collection | Arvéo |
|---|---|---|
| Platform metrics | Automatic | web01 CPU, DipAvailability of lbe-stNN-web |
| Resource logs | Diagnostic setting | Vault jobs, NSG counters |
| Activity log | Automatic (90 days), export at subscription level | Who changed what, when |
| Guest OS | Azure Monitor Agent + data collection rule (DCR) | Syslog and Perf of web01, web02 (M7) |
| Application | Application Insights | Out of scope (portal v3) |
Diagnostic setting = Categories (logs, metrics) + DestinationsAzureDiagnostics (shared) or resource-specific (recommended)az monitor diagnostic-settings categories list --resource "$VAULT_ID" \
--query "[].{Category:name, Type:categoryType}" -o tablesetting created at 14:00, query on 13:00 = empty table
Event = Operation + Caller + Timestamp + Status + ScopeQ="[].{Time:eventTimestamp, Operation:operationName.localizedValue, By:caller}"
az monitor activity-log list -g rg-st07-app --offset 4h --query "$Q" -o tableMetric query = Resource + Metric + Aggregation + Interval + PeriodVM_ID=$(az vm show -g rg-st07-app -n vm-st07-web01 --query id -o tsv)
az monitor metrics list --resource "$VM_ID" --metric "Percentage CPU" \
--interval PT5M --aggregation Average Maximum --offset 30m -o tableExpected output (values vary):
Timestamp Name Average Maximum
------------------- -------------- --------- ---------
2026-10-08 13:30:00 Percentage CPU 2.41 6.8
2026-10-08 13:35:00 Percentage CPU 1.97 3.12BackendIPAddress of DipAvailability)30 min
Alert rule = Scope + Signal + Condition + Severity + Action group
flowchart LR
P[Scope<br/>VM, group] --> S[Signal<br/>metric, log]
S --> C{Condition<br/>threshold, window}
C -->|true| A[Alert<br/>Sev0 to Sev4]
A --> G[Action group<br/>email, webhook]
[TO VERIFY] pricingRule type = Nature of the signal (+ latency + cost)| Type | Signal | Arvéo example |
|---|---|---|
| Metric | Platform or custom metric | CPU > 80 % on a VM of rg-stNN-app |
| Log search | Scheduled KQL query | 5 SSH failures in 10 min from the same IP |
| Activity log | Operation on a scope | Deletion of an NSG rule |
| Service Health | Azure incident, planned maintenance | Azure incident in France Central |
| Resource Health | Resource health (available, degraded) | web01 unavailable on the platform side |
Condition = Aggregation + Operator + Threshold + Window + Frequency[TO VERIFY] duration)Computer, BackendIPAddress)Before: CPU > 80 % over 1 min → alert on every spike (fatigue, ignored alerts)
After: average > 80 % over 5 min → sustained saturation only
Alert = Condition (Fired | Resolved) + User response (New | Acknowledged | Closed)flowchart LR V[Condition true] --> F[Fired<br/>notification] F --> R[Resolved<br/>condition false] F -.-> A[Acknowledged<br/>operator] A -.-> C[Closed<br/>operator]
Action group = Notifications + Actions (reusable by all rules)| Category | Examples | Arvéo use |
|---|---|---|
| Notifications | Email, SMS, push (Azure app), voice call | Operations team |
| Azure Resource Manager role | Email to holders of a role (Owner) | Subscription owners |
| Automation | Automation runbook, Function, Logic App | nginx restart |
| Integration | Webhook, ITSM, Event Hubs | Ticket in the support tool |
[TO VERIFY]Processing rule = Scope + Filters + Action + Schedulemonthly patching = scheduled processing rule, never disabled rules
Variables: <EMAIL> = mailbox address that receives the notifications
AG_ID=$(az monitor action-group create -g rg-st07-shared -n ag-st07-exploitation \
--short-name ag-st07-exp --action email exploitation <EMAIL> --query id -o tsv)
RG_APP_ID=$(az group show -n rg-st07-app --query id -o tsv)
az monitor metrics alert create -g rg-st07-shared -n alr-st07-cpu-vm \
--scopes "$RG_APP_ID" --target-resource-type Microsoft.Compute/virtualMachines \
--target-resource-region francecentral --condition "avg Percentage CPU > 80" \
--window-size 5m --evaluation-frequency 1m --severity 2 --action "$AG_ID" -o none40 min
Workspace = Tables + Retention + Pricing + Access controllog-stNN-shared (M3): Arvéo's single workspace, France Central, pay-as-you-go per GBHeartbeat, Syslog, Perf, AzureMetrics)[TO VERIFY] limitsdaily cap reached = collection suspended until the next day
Design = As few workspaces as possible + Justified exceptions| Criterion | One central workspace | Several workspaces |
|---|---|---|
| Cross queries | Simple (single scope) | workspace() or union, slower |
| Rights | Resource context, table RBAC | Clear separation per team |
| Sovereignty | A single region | One per data region |
| Cost | Pooled volume (commitment tiers) | Scattered volumes |
KQL query = Table | filter | transformation | aggregation | sort | renderHeartbeat
| where TimeGenerated > ago(1h)
| summarize Dernier = max(TimeGenerated) by Computer
| extend Minutes = datetime_diff('minute', now(), Dernier)
| extend Etat = iff(Minutes > 5, "SILENCIEUX", "OK")
| order by Minutes desc| (pipe): the output of one step feeds the nextwhere first: less data read, faster query==; =~ compares case-insensitively| Operator | Role | Example |
|---|---|---|
where |
Filter rows | where SeverityLevel == "err" |
project, extend |
Select, compute columns | extend Mo = Quantity |
summarize ... by |
Aggregate, group | summarize count() by Computer |
bin() |
Bucket time | by bin(TimeGenerated, 5m) |
parse |
Extract from text | parse SyslogMessage with * "from " Ip " " * |
join, render |
Join two tables, chart | render timechart (portal) |
Table = Source + Schema (+ ingestion delay)| Table | Fed by | Typical question |
|---|---|---|
Heartbeat |
Azure Monitor Agent (M7) | Which VM stopped responding? |
Perf, Syslog |
Rule dcr-stNN-linux (M7) |
web01 CPU, SSH failures |
AzureMetrics |
Diagnostic setting | Health of pool bp-web |
AddonAzureBackupJobs |
Vault setting (13:30) | Which backups failed? |
NTANetAnalytics |
Traffic analytics (S10.4) | Which flows to snet-web? |
Usage |
Workspace | Which table costs the most? |
Syslog
| where TimeGenerated > ago(1h) and Facility in ("auth", "authpriv")
| where SyslogMessage has "Failed password"
| parse SyslogMessage with * "from " IpSource " port " *
| summarize Echecs = count() by Computer, IpSourceWS=$(az monitor log-analytics workspace show -g rg-st07-shared -n log-st07-shared \
--query customerId -o tsv)
az monitor log-analytics query -w "$WS" --analytics-query "$(cat echecs.kql)" -o tableExpected output (technical column TableName omitted):
Computer IpSource Echecs
------------- ------------ --------
vm-st07-web01 203.0.113.50 8Log alert = Query + Measure + Dimensions + Threshold + Frequency + WindowComputer and per IpSource[TO VERIFY]test the query in Logs before turning it into an alert
Visualization = Workbooks | Dashboards | Managed Grafana | Power BIan "Arvéo Health" workbook shown at every operations meeting
25 min
Network Watcher = Diagnostics (one-off) + Monitoring (continuous) + Logging (flows)NetworkWatcher_<region> in NetworkWatcherRG, enabled with the first VNetNetworkWatcherRG (tools, flow logs)| Tool | Question | Extension on the VM |
|---|---|---|
| Next hop | Where does the packet go? | No |
| IP flow verify | Which NSG rule decides? | No |
| Effective security rules | Rules applied to the NIC? | No |
| NSG diagnostics | Verdict of each NSG on the path | No |
| Connection troubleshoot | Does the port answer? Where does it break? | Yes (NetworkWatcherAgent) |
| Packet capture | What does the traffic contain? | Yes |
| Tool | Use | Result |
|---|---|---|
| Connection monitor | Regular tests source → destination | Latency, loss, alerts |
| VNet flow logs | Each flow: 5-tuple, decision, bytes | JSON blobs (storage) |
| Traffic analytics | Flow aggregation and enrichment | Table NTANetAnalytics |
| Topology | Network resources and their links | Diagram in the portal |
fl-stNN-spoke-app created at 13:30, processed every 10 minDiagnosis = Route? → Filtering? → Listening? → Content?flowchart LR T[Ticket<br/>web01 → sql:1433] --> NH[Next hop<br/>route?] NH --> IP[IP flow verify<br/>NSG?] IP --> CT[Connection troubleshoot<br/>port open?] CT --> PC[Packet capture<br/>content?]
None: no point opening an NSG, the packet never leaves the subnetVirtualAppliance: ALSO check the firewall rules (outside Network Watcher)Flow log = Target (VNet, subnet, NIC) + Storage + Traffic analytics (optional)insights-logs-flowlogflowevent), adjustable retention[TO VERIFY][TO VERIFY] pricingflow log created during the incident = no trace of its start
web01 can no longer reach sql.arveo.internal:1433 (10.7.8.10)az network watcher show-next-hop -g rg-st07-app --vm vm-st07-web01 \
--source-ip 10.7.4.11 --dest-ip 10.7.8.10 \
--query "{Hop:nextHopType, Route:routeTableId}" -o table
az network watcher test-connectivity -g rg-st07-app --source-resource vm-st07-web01 \
--dest-address 10.7.5.100 --dest-port 8080 --protocol Tcp \
--query "{Status:connectionStatus, LatencyMs:avgLatencyInMs}" -o tableExpected output (latency varies):
Hop Route
------ ------------
None System Route
Status LatencyMs
--------- -----------
Reachable 1where, summarize ... by, bin(), parse; query tested before the alertweb01, then 2 %?web01web01 maintenance on Tuesday from 22:00 to 23:00: how to silence notifications?whereprojectsummarize ... byextendPerfHeartbeatAzureActivityUsageweb01 to 10.<OCT>.8.10 is None. Conclusion?nsg-stNN-data blocks the flow