Skip to content

Collector Release Notes

A collector installed from the current download can update itself once you turn that on for it. To move any other collector to a newer version, download the newer build from the Customer Portal → Downloads and re-run the installer on that host — it’s safe to run on an already-installed collector and won’t lose its enrollment. See Installing a Collector for the full steps, and Updates for how update delivery works overall.

Every published release is listed here, newest first. Entries are added as releases ship and are never removed, so this page is the running record from here on. It begins with the five most recent releases as of the day it was written; anything older predates the page.

For a tour of what GridNMS can do today, grouped by capability rather than by release, see What’s New.

  • A router that stops answering pings no longer silences the devices on your own network. Before reporting a device down, a collector checks whether it still has a path to the outside world, and holds back reports if it does not. The previous release narrowed that so an internet outage could not silence devices on your local network — but it made one exception: if the collector’s own router also stopped answering pings, nothing was reported at all. Plenty of routers simply do not answer pings, or stop answering under load, so that exception was reached far more often than a real outage warranted. The collector now judges from whether its own network connection is up, rather than from whether its router replies. A switch or a printer on the same network that is genuinely down now reports down in that situation, while devices the collector only reaches across the internet still hold their last known state. See Collector health.
  • A service check against a public address now reports “unknown” rather than “down” while the collector has no route to the internet. Previously every such check failed at once during an outage on the collector’s own connection, so a single dropped link produced an alert for each website, API and external service being watched — none of which had anything wrong with them. Checks against addresses on your own network are unaffected and keep reporting normally.
  • A check whose target actually answered still reports down. A refused connection or an unknown host name means something responded, so those are treated as a real failure rather than as a missing route — you still get alerted when a service is genuinely broken during an outage elsewhere. If any public target answers during a cycle, every held-back result in that cycle is reported as its check found it, so how long a check takes never changes the verdict. Where the collector cannot read its own network configuration completely, it reports rather than holds back: the behaviour leans toward telling you.
  • A device on your local network is no longer silenced when your internet connection drops. Before reporting a device down, a collector checks whether it can still reach the internet, and if it cannot it holds back those reports rather than alerting on everything at once. That check used to apply to every device the collector watched — including devices on the same local network, which an internet outage says nothing about. A printer or a switch that is genuinely down now reports down during an internet outage, while devices the collector reaches across the internet still hold their last known state. “On your local network” means an address the collector reaches without leaving its own link, including a device with a public address on that link — normal on an IPv6 network. If the collector’s own router stops answering as well, nothing is reported, as before. A collector on macOS cannot discover its router, so it always reports local devices rather than risk silence on a real outage; if such a host loses its connection entirely you may see a late batch of down-then-up events once it returns. See Collector health.
  • A collector on Linux now recognises an internet outage that leaves your local network working. Its connectivity check previously counted its own router as proof of connectivity, so an upstream outage with a reachable router read as healthy and every device the collector could no longer reach was reported down. Collectors on macOS were never affected. See Collector health for what the check now does.
  • Letting a collector scan a network on request no longer commits it to scanning on a schedule. Enabling port scanning and scheduling it are now two separate switches. Before this, the only way to run a one-off sweep of a network was to turn scanning on — and a network that is on is swept on a timer from then until you turn it off again, including once immediately each time the collector picks up new configuration. You can now permit a network and leave its schedule off, sweep it when you actually want to, and leave it quiet in between. Turning the permission off also clears the schedule, so switching it back on later does not silently resume a recurring scan you had stopped. Collectors older than this version do not understand a permit-without-schedule network and are simply not sent it; networks with both switches on behave exactly as they did.
  • One lost ping no longer marks a device down. A check used to send a single echo and wait three seconds, so a single dropped packet was reported as an outage — which happens routinely to a laptop on Wi-Fi, or to a host that was asleep. A check now sends several echoes, and a device counts as reachable if any of them comes back. Response time is the mean of the replies actually received; echoes that got no answer are left out of the average rather than counted as slow, so a host that spends its first packet waking up reports the speed it really answered at instead of an inflated one.
  • You can choose which DNS server a network’s host-name lookups go to. When a collector scans a network it asks DNS for the name belonging to each responsive host’s address, and until now that question always went to whatever resolver the collector’s own machine uses. You can now name a server on a network, a site, a collector, or the whole tenant; the narrowest one you have set wins, and leaving them all blank keeps using the collector host’s resolver exactly as before. If a network names a server that cannot be used, that network’s lookups stop and the reason is shown — rather than quietly asking a different server than the one you chose. Collectors older than this version do not understand the setting, so a server named for them had no effect.
  • The companion setting, if your hosts stopped showing names: looking up host names is now a switch of its own on the network, separate from port scanning, and it starts off. Turn it on for the networks whose discovered hosts you want named. The scan itself is unaffected either way.
  • An update that was rolled back now says so. A collector that updates itself checks each installed part against the published release after the swap, rather than trusting that it landed, and if it rolls the update back because the new version would not settle it tells the app. Before this, a collector that had reverted carried on reporting the version it was no longer running, and every status screen agreed with it.
  • A collector running in a container no longer claims it can update itself in place. A container is updated by pulling a newer image, and the image carries no updater — but it was reporting that it could update itself, so the app could offer an update that would never happen with nothing to explain why.
  • Nothing changes about what a collector collects, and there is nothing to re-configure. A collector with in-place updates turned on takes this on its own; on any other host, download the new build from the Customer Portal and re-run the installer.
  • A collector now reports the version it is actually running. An in-place update replaces several parts of a collector, and until now the version shown in the app was stamped when the update was prepared rather than read back from what ended up on the host. If a part of an update did not take — or the whole update was rolled back automatically after it failed to settle — the collector carried on reporting the newer version it was no longer running, and every status screen agreed with it. It reads the installed parts now, and an update that was rolled back says so in the app instead of looking like a success.
  • A log forwarder on this collector can receive every part of its own updater. A forwarder gets its updates through the collector it reports to. One part of the forwarder’s updater was not something the collector could pass along, so a forwarder that updated itself could end up with an updater missing a piece — quietly losing its own crash protection and its ability to update again. The collector carries every part now, each checked against the published release before it is handed over. This collector version is what makes the next forwarder release safe to install, so update your collectors before your forwarders.
  • Nothing changes about what a collector collects, and there is nothing to re-configure. A collector with in-place updates turned on takes this on its own; on any other host, download the new build from the Customer Portal and re-run the installer.
  • A rotated log-forwarder credential no longer locks the forwarder out of its control channel. When you rotate a forwarder’s credential there is a window in which both the old and the new one work, so the forwarder keeps talking while you apply the new credential on its host. Three of the four doors a forwarder uses already honoured that window; the control channel did not, which is why rotating a credential — or moving a forwarder to a different collector — could leave it refused on that one channel while everything else kept working and every status screen read healthy. It honours the window now. This collector version is what makes that window real, so a forwarder whose collector is older than 0.64 is told, in the product, that its control channel is not covered — rather than being quietly left unprotected.
  • Rotating a credential now asks whether it was exposed. If you answer yes, the previous credential stops working at once instead of staying valid for the window, and the new one is only ever shown to you — never made available for the forwarder to collect on its own. Answer no and the window behaves as before. Neither answer is pre-selected, because the right choice depends on something only you know.
  • Hosts get a name from DNS. When a collector scans a network you have turned port scanning on for, it now asks the DNS server its own machine already uses for the name belonging to each responsive host’s address, and the Hosts page shows it with a globe marker. Most of the hosts on a typical network have no name from any other source, so this is where their names come from. Nothing is sent to the host itself, lookups are capped and cached, and a DNS server that stops answering pauses lookups rather than slowing the scan.
  • Published for every platform: macOS on Apple Silicon and Intel, and Linux on 64-bit and ARM. The three releases before this one were Linux 64-bit only, so this is the first build macOS and ARM hosts have had since 0.59 — on those hosts it carries everything described under 0.60, 0.61 and 0.62 below, including the inspection-tolerant lane for networks that inspect outbound encrypted traffic.
  • On a Linux 64-bit host there is no functional change from 0.62.
  • Re-run the installer once on each host. It is safe on an already-installed collector and keeps its enrollment. That one run is what puts the current updater on the host; after it, a collector that has in-place updates turned on carries every part of itself forward on its own and needs no further manual step.
  • Fixes an in-place update replacing only part of the collector. The collector program was updated, but the smaller programs installed beside it were not — and one of those is what tells GridNMS which features the collector supports. So a host could finish an update running the new collector while still reporting the older feature list, which left newer settings refused for that collector until someone re-ran the installer on the host. An in-place update now carries every part, each one checked against our signature the same way the collector program already was.
  • Before it swaps anything, the collector now checks that each downloaded part actually runs, and refuses to swap if it cannot first back up what it is replacing. Previously a correctly signed but unusable part could be put in place, and the part that would have undone it was one of the parts being replaced.
  • Re-run the installer once on each host to pick this up. It is safe on an already-installed collector and keeps its enrollment. After that, in-place updates carry every part on their own.
  • Published for Linux on 64-bit only. macOS and ARM hosts stay on 0.59.
  • Adds support for networks that inspect outbound encrypted traffic. On such a network a collector could not previously send anything to GridNMS at all: the inspecting equipment stands in the middle of the connection, so the certificate the collector uses to prove which collector it is never reaches us, and everything it sent was refused. A collector on 0.61 can prove its identity inside each request instead, which inspection leaves intact.
  • This is off until you turn it on, per collector, under Collectors → the collector → Settings → Encrypted-traffic inspection. Leave it off unless a collector is on a network that inspects traffic: every other collector keeps using the stronger connection-level proof, and nothing about those collectors changes in this release.
  • Worth knowing before you turn it on: on this setting a collector sends over a connection your network’s equipment can read, so the equipment can see the monitoring data in transit. That is the trade for being able to send at all from such a network.
  • Published for Linux on 64-bit only. macOS and ARM hosts stay on 0.59.
  • Fixes an in-place update that had already failed a few times never being retried, even after the reason it failed was gone. A collector that had given up stayed given up, and the only way back was a command on the host. Asking for an update from GridNMS now clears that and lets the collector try again, while the brake that stops a collector restarting in a loop still works as before.
  • When our side refuses an update, the collector now records what it was told instead of reporting that nothing answered at all. Previously a refusal and an unreachable server looked identical in the log.
  • A collector now checks for an update about a minute after it starts, rather than waiting a full interval, and notices a new update straight away instead of on its next scheduled check.
  • Fixes a collector installed directly on a macOS host switching its own periodic update check off silently. It affected every such collector, whatever its version.
  • Published for Linux on 64-bit only. macOS and ARM hosts stay on 0.59, which is current for them; the fixes above that matter on macOS arrive with its next build.
  • Published for every platform from a single build: macOS on Apple Silicon and Intel, and Linux on 64-bit and ARM. The macOS download had been several releases behind; this brings it level with Linux.
  • macOS hosts moving up from 0.55 also pick up the release-signature check added in 0.56 and the 0.57 fix for a collector reporting some of its own details as blank — the list of features it supports, its local IP address, its local DNS name, and its network interfaces.
  • No functional change for Linux hosts already on 0.58.
  • No functional change. This release exists so that the in-place update mechanism can be exercised end to end on a real host: the collector it replaces is 0.57, and everything it runs is the same. If you install it directly you will notice nothing different.
  • In-place collector updates are not switched on for anyone yet. Until they are, moving a collector to a newer version is still a matter of downloading the newer build and re-running the installer on that host.
  • Fixes a collector installed directly on a host (not in a container) starting up with none of its receiving ports open, while the host’s service manager and GridNMS both still showed it as healthy. It affected a fresh install and every restart afterwards, and the only sign was that nothing arrived.
  • Fixes a collector installed directly on a host reporting some of its own details as blank — the list of features it supports, its local IP address, its local DNS name, and its network interfaces. Which details were blank depended on when that collector was first installed rather than on its version, so re-running the installer had not previously picked them up. Re-run the installer on the host once to pick this up.
  • Groundwork so that a future in-place update can hand its receiving ports to the new version without dropping anything in between. Nothing changes for you in this release.
  • Every collector release we publish is now cryptographically signed. The check that refuses an update whose signature doesn’t verify is part of the in-place update mechanism below, so it starts protecting a collector when we switch that on for it. Updating by re-running the installer is unchanged.
  • Groundwork for updating a collector in place, without re-running the installer on the host. It ships switched off: nothing updates itself yet, and upgrading a collector is still a matter of re-running the installer. We enable it per collector once it has been proven.
  • Fixes a case where a collector installed directly on a Mac (not in a container) left some of its own details blank: its local IP address, local DNS name, network interfaces and routes, and the list of features it supports. Container installs were not affected. Reinstall the collector on each Mac host to pick up the fix.
  • During a log forwarder credential rotation, the collector now keeps accepting the previous credential until the forwarder confirms it has switched to the new one, so a rotation can’t interrupt log delivery.
  • Adds Run now for command-based (SSH) collection, so you can trigger a collection on demand instead of waiting for the next scheduled run.
  • The configuration-backup command is now set per device class, instead of assuming a Cisco-style command for every device.
  • Internal fix to how the collector reports one of its own relay capabilities. No operator-visible change.
  • Adds SSH host-key pinning for command-based collection: the collector checks a device’s SSH host key against a pinned fingerprint and refuses the connection if it doesn’t match, instead of trusting it silently on first use. Older collectors ignore the setting, so you can roll it out gradually across a fleet.
  • Supporting work for log forwarder self-upgrade delivery. No operator-visible change on the collector itself.

docs built 2026-11-12 · 945a2dec2