Skip to content

Latest commit

 

History

History
628 lines (520 loc) · 35.1 KB

monitoring.md

File metadata and controls

628 lines (520 loc) · 35.1 KB

Monitoring and alerts

Since 0.29, Rem0te watches the computers it manages and raises an alert when something needs a person: a computer offline, a disk nearly full, a Windows service stopped, matching events in an event log, a restart pending for days. Alerts are deduplicated, a site-wide outage is recognised as one problem, alerts can be held for maintenance, and unacknowledged alerts escalate to further destinations — including a ticket in ClientSt0r.

Monitoring only reads. A monitor never changes a computer. Everything it uses is something the agent already reports: the heartbeat (CPU, memory, disk, restart pending), the inventory, and a small monitor report listing the services and event queries the server asked about.

Monitors judge each reading as it arrives. The same readings are also kept over time (0.33) and charted on a computer's Performance tab — see metric-history.md. History raises no alerts, and monitoring does not read it.

Alerts page: severity totals, the filter, and an expanded site-outage alert with its history

  1. Alerts in the sidebar — shown to anyone with monitoring:alerts.
  2. Totals for what is open now, including how many are held by a maintenance window.
  3. Filter by state and severity. The default is open and acknowledged.
  4. A site outage lists the computers it gathered; they are not notified on their own while it is open.
  5. Acknowledge stops escalation; Resolve needs a note.

Where things are

What Where Who
Alerts Alerts in the sidebar anyone with monitoring:alerts
One computer's alerts, monitors, change history and software Computer → Monitoring tab anyone who can see the computer
Monitors, maintenance windows, notifications Settings → Monitoring & API → Monitoring monitoring:configure for the whole business, a Business Owner, or the MSP

Presets that allow viewing inventory also grant monitoring:alerts. A technician sees alerts only for computers their grant covers; a site-outage alert needs the whole business in view.

Monitors

A monitor has a key, a kind, a severity (info, warning, critical) and parameters.

Kind Raises when Clears when
Computer offline no heartbeat for minutes (default 15) the computer reports again
Disk space system disk below minFreePercent free two good heartbeats in a row
Processor busy / Memory used above percent for samples heartbeats in a row two good heartbeats
Windows service stopped the named service is not Running (optionally: missing; disabled ignored by default) it is Running again
Event log match at least threshold events since the last report match ids, levels or provider in one of System, Application, Security, Setup no match for clearAfterMinutes
Restart pending Windows has wanted a restart for hours the restart happens
Agent silent (0.36; recommended from 0.36.1) RustDesk still reaches the computer but its agent has not checked in for minutes (default 60, 30 – 10,080) the agent checks in again
Drive space, every fixed drive (0.55; recommended) any fixed drive (C:, D:, …) the inventory reports is below minFreePercent (default 10) or minFreeMb — one alert per drive that drive is 2 points above the line, or is no longer reported
Not restarted (0.55) the computer has been up for days (default 30, 1 – 365) it restarts
Security updates outstanding (0.55; recommended) a security or critical update has been offered for days (default 30, 1 – 180) it is installed, or no longer offered
Update install failed (0.55; recommended) an update failed to install in the last days (default 7, 1 – 90) it installs, or the failure is older than the window

An event-log monitor must name at least one of IDs, levels or provider — a monitor on every event in a log is a flood, not a monitor. The agent asks for at most 100 events per query and sends the count and the newest event (its ID, level, provider and the first 300 characters of its message); from 0.66 it also sends the five newest, so each can be classified (see Alert classification below). An agent older than 0.66 keeps working with one.

Add recommended creates a starter set — offline, disk, CPU, memory, restart pending, disk errors (System log IDs 7, 11, 51, 52, 55 and 153 at levels Critical, Error and Warning; before 0.66 IDs 7, 11, 51 and 153 at any level — see below), from 0.36.1 agent silent (60 minutes, warning), and from 0.55 other fixed drives (10 %, C: excepted), security updates outstanding (30 days) and failed update installs (7 days), all warning — as ordinary monitors you then adjust. Nothing is created without that click: no monitor is installed at startup, by an upgrade or by the seed, and the click adds only the starters missing at that scope. A scope where someone pressed Add recommended before 0.36.1 has no agent-silent monitor until someone presses it again or adds it by hand.

Agent silent (0.36)

A computer can be reachable without reporting. isOnline stays true while hbbs sees its RustDesk peer, so a computer whose agent died but whose RustDesk still runs shows amber Reachable (RustDesk) · agent silent on its page (0.35, architecture.md) — and its monitoring, patch and inventory values stop updating. The Offline monitor cannot see this: it watches isOnline, which is true. The agent-silent monitor raises when:

  • isOnline is true (RustDesk reaches it), and
  • the agent's last heartbeat is at least minutes ago (default 60).

It is never on by default, and nothing creates it for you. Since 0.36.1 it is one of the monitors Add recommended creates (the owner's decision; in 0.36 it was offered only on its own); you can also create it alone under New monitor → Agent silent (RustDesk still reachable), which fills in the key agent-silent, 60 minutes and warning. It follows the usual scopes — global, business, site, device group (by priority) and computer — so you can turn it on for one business, give kiosks a longer threshold, or switch it off for a site.

  • Threshold 30 minutes to 7 days. Not below 30 minutes: that is the window in which the page still says Online, and an alert would contradict it.
  • The alert reads "PC: agent silent for 3 d — remote control should still work", and its detail says what to do: Update agent on the computer's Agent card, or re-run the installer on the machine. The title's age is rewritten as it grows (tens of minutes, then hours, then days), not every sweep.
  • Clears on the first sweep after the agent checks in.
  • Not reachable any more — the computer drops off RustDesk too: no new agent-silent alert is raised (that is the Offline monitor's case), and an open one stays open rather than resolving and re-raising every time hbbs restarts.
  • Never reported — a computer whose agent has never checked in (RustDesk only, or an install that never completed) is not this monitor's case.
  • Where it shows — the alert list, the computer's Monitoring tab, an Agent silent alert chip at the top of the computer's page, and a line on its Agent card, the last two linking straight to the alert.
  • Not sent to ClientSt0r. The integration contract lists the alert kinds it carries and this is not one; with ticketing on, an agent-silent alert is notified as usual but not offered for a ticket. Carrying it needs a contract revision agreed with ClientSt0r.

Drives, uptime and updates (0.55)

These four read only what the server already stores, so they need no agent update. The sweep looks at them at most every five minutes (MONITOR_INVENTORY_SECONDS, default 300).

  • Drive space (every fixed drive). This reads the drive list from the inventory pass. On Windows that is every Win32_LogicalDisk of type 3 (a fixed drive), and the pass runs about every six hours. A pass older than 48 hours is not used. Each drive gets its own alert, keyed <monitor>:<computer>:<letter>, so D: and E: filling up are two problems. Only a single letter and a colon counts as a drive. A Linux or macOS mount point is ignored, and so is a malformed entry.
    • Only these drives narrows the monitor to the drives you name.
    • Except these drives leaves drives out.
    • How it relates to Disk space. The older Disk space monitor reads the system disk from every heartbeat, so it is quicker. The recommended drive-space monitor therefore leaves C: out, so a full C: is not alerted twice.
    • It clears only at the line plus 2 points, so a drive hovering on the line does not flap.
    • A drive that disappears from the inventory, such as a removed disk, resolves its alert.
  • Not restarted. This uses the boot time from the computer's heartbeat. An offline computer is not judged, because it may have restarted since. This monitor is not in the recommended set.
  • Security updates outstanding. This needs the signed agent's patch scan. On a standard agent, or with a scan more than 36 hours old, the monitor does nothing.
    • An update's age counts from when Rem0te first saw it offered. Windows does not say when an update was released. On a computer that is new to Rem0te, the clock starts at its first scan.
    • Only the Security and Critical classifications count. A browse-only (optional) update never does.
  • Update install failed. This reads Windows Update's own install history from the patch scan, and the results of Rem0te patch deployments to that computer.
    • A failure followed by a later successful install of the same update is forgotten.
    • The alert lists up to ten failures with their KB and error code.

Automatic tickets (0.55)

Settings → Monitoring → Automatic tickets opens a ticket in the PSA when an alert opens. You set this at the platform level, and a business can have its own rule. A business's rule replaces the platform rule there, even when it is switched off. The rule says:

  • the lowest severity that opens a ticket (default critical);
  • optionally, which kinds open one (none ticked means every kind).

The ticket goes through the PSA connector under Settings → PSA / ticketing: the business's own, or else the platform default. It is the ticket Create ticket on that alert would make: the alert's title, kind and severity, the computer, the business and a link back to the alert. It carries no alert detail, no note and no AI text (see ai-and-tickets.md). It is marked as requested by Rem0te monitoring (automatic).

When an alert is ticketed:

  • Once. There is never more than one automatic ticket per alert; the database holds one row per alert.
  • Only alerts that open after the rule is switched on. Switching it on does not ticket the backlog.
  • After the alert's notification hold. An Offline alert waits for the site-outage correlation hold, as its notification does.
  • Never for an alert in a maintenance window, for a site outage itself, or for a computer gathered under an open site outage.
  • Never under a notifications containment pause. An alert held by a pause is ticketed when the pause is lifted, if it is still open.
  • At most 10 sends per business an hour (PSA_AUTO_TICKETS_PER_HOUR). An alert over the budget waits for a later sweep.

When a send fails:

  • The failure is recorded on the alert's history and in the audit log.
  • It is tried again after 5, 30 and then 120 minutes, four sends in all. After that it is not sent.
  • A send interrupted by an API restart is given up rather than repeated, because the PSA may already have made the ticket.
  • It stops being retried once the alert resolves or the rule no longer asks for it.

Where the ticket shows: the reference, with the PSA's link when the PSA returned one, appears on the alert in Alerts, on the computer's Monitoring tab and in the dashboard's Recent alerts. Resolving the alert does not close the ticket, because no connector supports that yet.

Who may change the rule, and what is recorded:

  • The rule follows who is notified (below). The platform rule belongs to the MSP. A business rule belongs to the MSP, that Business Owner, or someone with monitoring:configure for the whole business.
  • Saving or removing it asks you to confirm it is you (monitoring.autoticket under Settings → Confirm it is you).
  • Each change is audited (MONITOR_TICKET_POLICY_CHANGED, before and after).
  • Each send is audited before it leaves (PSA_AUTO_TICKET_REQUESTED), then as PSA_TICKET_CREATED or PSA_TICKET_FAILED. These entries carry ids, the kind, the severity and the reference, never the subject or the body.

The pass runs inside the monitoring sweep, so MONITOR_SWEEP=off stops it.

Monitoring settings for one business: its own monitors above the inherited global ones

  1. Business — the MSP picks one; blank is the global layer every business inherits. A Business Owner sees only their own.
  2. Monitors, maintenance windows and notifications are separate tabs.
  3. Add recommended — the starter set, added as ordinary monitors.
  4. Overridden below — this business has its own monitor with the same key, so the global one does not apply here.
  5. Switch off for this business — adds a disabled override with the same key.

Inheritance

Monitors are set at five levels: global (the MSP, for every business), business, site, device group (0.35) and computer. For each key, the most specific monitor wins — computer, then group, then site, then business, then global. So to change the offline threshold for one business, add a business monitor with the key offline; to switch it off there, add it disabled ("Switch off for this business" does exactly that). Removing the override brings the inherited one back.

A device-group monitor applies to the computers in that group now — a rule group's members are computed at each evaluation — and carries a priority: a computer in two groups with a monitor of the same key follows the lower number (1 first), and two group monitors with the same key cannot share one. A group that picks computers by their alerts cannot carry a monitor. See Device-group policies for the full rule, what the pages show and the measured cost.

A computer's Monitoring tab leads with its alerts (open ones first, worst severity first; resolved ones folded away), then lists the monitors actually in effect — by name and kind, with what each watches for in words — and where each came from. Its open alerts also show in the attention strip at the top of the computer's page. From 0.31 installed software has its own Software tab.

A computer's Monitoring tab: its open alerts first, the monitors in effect, and change history

  1. The Monitoring tab on a computer's page (?tab=monitoring links straight to it).
  2. Its open alerts, first.
  3. Monitors in effect, where each came from, and what the agent last reported for services and event queries.
  4. Changes between inventories. Installed software is on the Software tab (?tab=software) from 0.31.

Alert classification (0.66)

Every event an event-log monitor matches is judged before anything is raised, by one deterministic, versioned classifier (apps/api/src/monitoring/classify.ts). It never needs AI, and it gives the same answer for the same event every time.

Class Means Alert
Critical a catalogued event that strongly indicates a failure (disk bad block, controller error, the disk's own SMART prediction, NTFS corruption) critical
Warning a catalogued event worth investigating (paging error, I/O retried, a device reset) warning; critical when it recurs 5 or more times in one report
Informational Windows logged it as Information or Verbose, and the monitor did not ask for that level info
Benign a known harmless pattern, matched on every structured field none: the event is recorded, not raised
Unknown Rem0te does not know the event the monitor's own severity

The rules that keep this safe:

  • Never above the monitor. The severity you set on a monitor is the ceiling. The classifier can lower an alert only on positive evidence: a known-harmless pattern, a catalogued event, or an Information/Verbose event under a monitor that did not ask for that level.
  • An event Rem0te does not know keeps the monitor's severity. Your own monitors (a backup product's failure event, an application's error) are never lowered.
  • The Security log is never lowered by level, and neither is level 0 (LogAlways), which is what audit events use.
  • Counted but not seen. An agent older than 0.66 sends only the newest of the matching events. When the agent counted more than it sent, the rest cannot be called harmless, so the alert stays at least warning.
  • A harmless event cannot hide a real one. A report holding a harmless event and a disk error raises the disk error; the harmless one is only taken out of the count the monitor's threshold sees.
  • Known-harmless patterns are narrow. Each names the log, provider, event ID, level (Information or Verbose only), an 8-digit status code and the affected component, and every one of them must match. The status code and component are read from the device path and the 0x… code in the message, which Windows does not translate — never from words in the message.

The first known-harmless pattern: TxR inside a shadow copy

Microsoft-Windows-Kernel-General event 11, level 4 (Information), status 0xC00000A2 (STATUS_MEDIA_WRITE_PROTECTED), on a \Device\HarddiskVolumeShadowCopy<n> path: "TxR init phase for hive … finished with result=0xC00000A2". Windows tried to start the transactional registry for a hive inside a Volume Shadow Copy, which is a read-only snapshot. Backup and System Restore software causes it. It is not a disk error. Any shadow-copy number matches; another provider, level, status code or component does not. Event ID 11 is not suppressed: disk 11 and every storage controller driver's own 11 (atapi, iaStor…, storahci, stornvme, LSI, megasas…) are critical.

Why the disk-errors monitor changed

Before 0.66 the recommended disk errors monitor watched IDs 7, 11, 51 and 153 in the System log from any provider at any level. On 2026-11-18 it raised a critical disk alert on DS-HOME1 for the TxR event above. From 0.66 it asks the agent only for levels Critical, Error and Warning (the genuine disk events are Error or Warning), and adds 52 (the disk's SMART failure prediction) and 55 (NTFS corruption). It deliberately names no provider: "controller error" (11) is logged by each storage controller driver under its own name, and naming only disk would have dropped them. The classifier then judges each event by provider.

On upgrade, the API corrects every stored disk-errors monitor whose parameters are still exactly the old recommended ones, and records it in the audit log (Monitor changed, op: corrected, before and after). A disk-errors monitor whose parameters someone changed is left alone — review it under Settings → Monitoring.

Open event alerts raised before 0.66 are classified again on start from the event they stored. An alert's severity moves to the classifier's; one whose every counted event was a known-harmless pattern is resolved by the monitor with a note. Each change is in the alert's history (Reclassified) and the audit log. Nothing is deleted.

What is not monitored

Rem0te reads only what the agent's event-log monitors and inventory report. There is no SMART polling (only the disk's own event 52, if the monitor asks for it), no backup-product integration (a backup failure is monitored only if you add an event-log monitor for that product's event), and no RAID-controller health beyond the Windows event log. One event is never treated as proof that hardware is failing; the explanation says so.

Event fingerprints

Every event the agent sends gets a fingerprint: its log, provider, event ID, level, status code and component, and its message with what varies taken out (shadow-copy numbers, GUIDs, dates and times, process IDs, temporary paths, user-profile names, block addresses and long hexadecimal values). Status codes, disk numbers and internal codes stay. Each computer, monitor and fingerprint is one row (MonitorEventFingerprint), counted every time it is reported, with the newest raw event — including the harmless events that never became an alert, so nothing reported is lost. View similar alerts on an alert lists the alerts with the same fingerprint that you can see.

What an alert says

Every alert has a plain-English explanation: What happened?, Why was this alert generated? (the monitor, what it watches, and how the severity was decided), How serious is it? (honest about what one event can and cannot show), Recommended action, and Classification details (a known pattern, a catalogued event, the event level, or "unverified"). Under it: View raw event, View similar alerts and View matching monitor rule. The alert list shows a small badge for an event alert's class (known failure, known warning, informational, unverified, partly checked).

False positives and alert quality rules (0.67)

An alert that was not a real problem is reported, not silenced. Fixing the cause is a rule, which is proposed, tested and approved before it changes anything. The two are kept apart on purpose.

Report false positive

On any alert (except a site outage), Report false positive shows the alert, its monitor and classification, the original event, how many similar alerts you can see and on which computers, and asks why:

Known harmless event, Incorrect severity, Duplicate alert, Expected activity, Incorrect threshold, Monitoring rule too broad, Incorrect correlation, or Other (which needs a comment).

  • Also resolve this alert resolves that one alert, nothing else.
  • Recommend preventing similar alerts in the future adds the report to a recommendation for that monitor and event signature in that business. It shows the proposed handling as a suggestion.
  • A report changes no monitor and creates no rule. Anyone who can see the alert can report it, once. The report is in the alert's history and audited (ALERT_FALSE_POSITIVE_REPORTED, without the comment).

Recommendations

Alerts → False positives & rules lists open recommendations: the issue, the recommendation, its impact and a caution. More reports, from more people and on more computers, raise its priority. They never approve anything. Preview matching events tests the suggested rule against stored alerts without saving it. Edit scope and propose turns it into a proposed rule; Dismiss closes it with a note.

Alert quality rules

A rule names:

Scope one computer, a device group, a business, one monitor, an operating system, or every business (platform, Platform Admin only). It never reaches outside its business.
Which alerts the monitor key and, for an event-log monitor, the event fingerprint or both the provider and the event id (plus optional levels, status code, component). A rule on a monitor's every event is refused (RULE_TOO_BROAD). A threshold monitor (disk space, offline…) takes no event conditions and cannot be changed everywhere at once.
What it does suppress (the event is still recorded, no alert), lower to warning or info, alert only at a count in one report, or alert only after repeated reports.
How long temporary (ends by itself, at most 90 days) or permanent with a review date (at most a year).

A rule only ever makes an alert quieter: it never raises one above the monitor's severity, and the classifier's own rules (0.66) still apply first.

Precedence. When two active rules match, the narrower scope wins (computer, group, business, monitor, OS, platform); then the one with more conditions; then the one that keeps more of the alert. There is one active rule per scope and conditions (RULE_CONFLICT).

Proposing and testing. A proposal changes nothing. While pending, a rule is evaluated in shadow: it is counted on matching alerts but not applied. Test proposed rule replays it over the last 90 days of stored alerts (at most 5,000) and the recorded event signatures, and shows what would have been suppressed or lowered, on which computers and businesses, and any dangerous matches: events the classifier knows as genuine failures, or Security-log events.

Approving. Approval needs a test of the current revision less than seven days old (SIMULATION_REQUIRED), an explicit acceptance of any dangerous matches (DANGEROUS_MATCHES), and someone who may configure monitoring for that scope. A high-risk rule — one that lowers a critical monitor, or touches the Security log, a backup monitor, storage events, every business, or has dangerous matches — needs a second, different approver (SECOND_APPROVER_REQUIRED). Approving may ask you to confirm it is you (Settings → Confirm it is you, Approving alert quality rules, off by default). On approval, open alerts it covers are re-checked at once, each change recorded as Reclassified.

Rolling back, revising, expiring. Roll back now ends a rule at once; if it had replaced an earlier version, that version is restored. Revise makes a new proposal that replaces the active rule only once approved; the old version is kept. A temporary rule expires in the monitoring sweep; a permanent one past its review date is flagged review overdue and keeps working. Every step is in the rule's history and the audit log (ALERT_QUALITY_RULE_PROPOSED, _EDITED, _APPROVED, _REJECTED, _ROLLED_BACK, _EXPIRED).

Notifications and tickets. Rules apply before an alert is raised, so a suppressed alert notifies nobody and opens no ticket. A lowered alert is notified at its new severity. Automatic tickets are also de-duplicated: the same event signature on the same computer within 24 hours links to the earlier ticket instead of sending a new one.

Who hears about it. A proposal, approval, rollback or expiry is mailed to confirmed recipients: Platform Admins for a platform rule, the business's owners for a business's. A notifications pause stops the mail.

The page

Alerts → False positives & rules also shows totals (reports, pending recommendations, rules awaiting approval, approved rules, temporary exceptions, expiring in 7 days, review due, alerts prevented), the most reported monitors, computers, businesses and reasons, a 12-week accuracy trend (the share of alerts nobody reported — unreported is not the same as confirmed), filters, and monitoring-quality checks: monitors with a high volume, flapping, never triggering, duplicated, with a sensitive threshold, escalating informational events, often reported, or alerting after an agent update; conflicting rules; rules that suppressed known failures; and rules due for review. The checks are recommendations only.

AI

The alert's Ask AI (Settings → AI) is advisory only. It can help read an event; it never reports, proposes or approves anything. With no AI configured, everything above works the same.

Limits

  • A business with one person who configures monitoring cannot approve a high-risk rule: it needs a second approver.
  • A backup monitor is recognised (for risk) by its key, name or provider containing "backup".
  • The agent-update check uses each computer's current agent version.
  • A threshold rule is not shadow-counted while pending; the test covers it.
  • Alert only after repeated reports is approximated in the test from the stored alerts.

Alerts

There is one alert per problem, not per occurrence: while an alert is open, a repeat raises its count (×3) and moves last seen. When the monitor sees the problem clear, the alert resolves itself ("resolved (cleared)").

  • Acknowledge — "someone has this." Stops escalation. A note is optional.
  • Resolve — closes it with a note (at least three characters). If the problem is still there, the next check raises a new alert.

Offline alerts wait three minutes before anyone is notified, so a computer that restarts does not page anybody.

Site outages

When three or more computers at one site are offline and they are at least half of that site, Rem0te raises a single critical site outage alert and groups the offline alerts under it. While it is open the grouped alerts are not notified on their own — one page for a site that lost its internet, not forty. It clears when fewer than a quarter (or fewer than two) remain offline; the grouped alerts still open are then notified normally.

Sites need at least three computers for this to apply.

Maintenance windows

A maintenance window holds notifications for a business (or every business), optionally only for some monitor keys, for up to 14 days and scheduled up to 90 days ahead. Alerts are still raised and shown, marked maintenance; anything still open when the window ends is notified then. Ending a window early releases what it held at once.

A maintenance window is not containment. It changes who is told, not what reaches a computer. To stop changes reaching computers during an incident, use containment.

Who is notified

Settings → Monitoring → Notifications holds up to four escalation levels. Each fires once an alert has been open (and unacknowledged) that many minutes, for alerts at or above its severity, and sends to each of its destinations (up to five): a JSON POST to a webhook (Slack- or Teams-style incoming webhooks, a ticketing relay, your own endpoint), or, from 0.34, an email to a confirmed recipient — picked from the business's (or the global) list of addresses whose owners clicked a confirmation link, never typed. The email carries only the alert's title and severity, the computer (or site), the business and a link; see Email and scheduled reports. A level is fired once per alert; when the alert resolves, every level that was notified is told.

{ "type": "alert.notify", "level": 0, "alert": { "id": "…", "title": "PC-12 has been offline for 15 minutes", "severity": "warning", "state": "open", "business": "…", "computer": "PC-12", "firstSeenAt": "…" } }

Notifications: two escalation levels, the ClientSt0r ticket option, and Save

  1. Level 1 fires at once for warnings and above.
  2. Level 2 fires after 30 minutes unacknowledged, for critical alerts only.
  3. Offer alerts at or above a severity to ClientSt0r for a ticket.
  4. Saving asks you to confirm it is you.

A business uses its own policy if it has one, otherwise the global one.

  • A business's policy is set by a Business Owner of that business or the MSP; the global one by the MSP.
  • Saving asks you to re-authenticate: a destination receives alert text, including computer names.
  • A business destination must be https on a public host name — no IP addresses, localhost or internal names. The audit log records hosts only, because a webhook URL's path is often its secret; people who cannot change the policy see addresses shortened.
  • An email destination must be a confirmed recipient of the policy's own scope, checked when the policy is saved and again when mail is sent. A notifications containment pause stops alert emails; webhooks keep going.

Tickets in ClientSt0r

With Offer alerts to ClientSt0r on, alerts at or above the chosen severity are sent to ClientSt0r as signed alert events (alert.raised, alert.updated, alert.resolved). alert.raised is sent when the first notification would be, so a maintenance window holds it as well. ClientSt0r decides whether to open a ticket and links it back; the alert then shows the ticket. A restart pending alert also carries a redacted patch summary (0.30) for the ticket. Details and the contract routes are in the ClientSt0r integration.

Change history and software

Every inventory is compared with the one before. Changes to hardware, operating system, disks, network adapters and installed software are kept on the computer's Monitoring tab — "software added: 7-Zip 24.08", "memory changed: 16 GB → 32 GB". The first inventory is the baseline and records no changes.

Installed software comes from the Windows uninstall registry (64- and 32-bit), skipping system components and updates, up to 500 entries. Agents older than 0.29 do not report it; the tab says so.

Operator settings

Variable Default Meaning
MONITOR_SWEEP on off stops the sweep (offline, agent silent, restart pending, outage correlation, notification)
MONITOR_SWEEP_SECONDS 60 how often the sweep runs
MONITOR_OFFLINE_HOLD_SECONDS 180 how long an offline alert waits before notifying

Troubleshooting

  • No alerts at all — check Settings → Monitoring has monitors (add the recommended ones), and that the computer's Monitoring tab lists them.
  • A service monitor never fires — the service name is the short name (Spooler, not Print Spooler). The tab shows what the agent reported.
  • Nobody was notified — the alert's history shows each notification and its HTTP result. Check for an active maintenance window, an acknowledged alert, or a level whose minimum severity is above the alert's.
  • A computer shows "agent silent" but no alert — the agent-silent monitor exists only where someone added it — with Add recommended (0.36.1) or under New monitor. The Offline monitor does not fire for it, because RustDesk still reaches the computer.
  • A disk-errors alert that is not about a disk — open it and read Classification details. Unverified means Rem0te does not know that provider's event: the alert has the monitor's severity. If the event is harmless, resolve the alert and narrow the monitor (levels or a provider). A disk-errors monitor someone edited before 0.66 still has its old parameters — compare it with the recommended ones above.
  • An alert was not raised although the event is in the log — open the computer's Monitoring tab: the event's signature is recorded. Then Alerts → False positives & rules: an active rule for that scope may suppress it. An alert a rule changed says changed by a rule and names it.
  • Approve is greyed out — the rule has not been tested, or the test is for an older revision or more than a week old: Test proposed rule again. A high-risk rule you approved first waits for someone else.
  • RULE_TOO_BROAD — name the event fingerprint (or the provider and event id), or scope a threshold rule to a business, group or computer.
  • An alert says "partly checked" — the agent counted more matching events than it sent (agents older than 0.66 send one). Look in the computer's event log at the time of the alert for the others.
  • Forty offline alerts for one site — the site needs at least three computers assigned to it for outage grouping to apply.