Questions and troubleshooting¶
New in the next release
This is on the dev and qa builds now, and in the next release.
What you see, and why¶
| What you see | Why, and what to do |
|---|---|
| No Capacity tab | Ferrule watches no hypervisor or storage source yet. Add one |
| The tab says capacity planning is part of Business and MSP | No license or trial in force. History is still being kept |
| The tab says The Capacity tab couldn't be drawn | Something on the page failed. Report it, with the error from the browser's console (F12) |
| needs N more days of history | Under three weeks of history since the last change of capacity. Wait; see why |
| A row that had a date says it needs more history again | Its capacity changed by more than 2%: a host added or removed, a datastore grown, spare hosts set. The forecast restarts from that day |
| not rising | Its use is flat or falling. Good news |
| not within a year | It's rising, but too slowly to reach its threshold in 12 months |
| A forecast date jumps around | The trend is noisy; look at the range. Disk forecasts settle as more days come in |
| No vCPUs allocated or Memory allocated rows | They come from the inventory. Run one: Virtualization → Inventory → Take inventory now |
| A cluster's CPU looks low on vCenter | CPU is counted in cores used, not megahertz. 5% of a 32-core cluster is 1.6 cores |
| History shows (N from inventory) | Those days were read back from old inventories: disk and allocation only |
| A backup store's row is amber with "retention N% filled" | Its retention is still filling, so its date may be early. See Storage |
| A backup store shows no policy | Neither PBS nor any Proxmox VE backup job writing to it sets one (or it's set to keep everything), or the PVE token can't read jobs: give it PVEAuditor |
| dedup unknown: can't read garbage collection | The PBS token can't read that datastore's garbage-collection status (it needs the Audit role there), or garbage collection has never run |
Planning¶
| What you see | Why, and what to do |
|---|---|
| How many more fit shows 0 in red | The next VM of that size crosses a threshold. The note says what stops it |
| Move VMs says no VMs seen yet | VMs are listed from the first poll after this release. Wait a poll |
| Add a host says host sizes come with the next day's summary | The cluster's largest host is learned at midnight. Try again tomorrow |
| Add disks doesn't list a pool | Its NAS doesn't report its layout. See what each reports |
| Adding disks gives less than their size | That's the layout: RAIDZ keeps parity, SHR keeps the largest disk, RAID 5 counts the smallest. The note explains |
| Adding disks gives nothing | Fewer disks than a vdev is wide, or RAID 1. The note says what to do instead |
The rack check¶
| What you see | Why, and what to do |
|---|---|
| No inventory yet | It learns each host's NICs from the inventory. Take one |
| Can't see where the hosts plug in | No switch Ferrule reads has the hosts' or their VMs' addresses, and no LLDP names them. Add the switches, or turn on LLDP. See where hosts plug in |
| has N more links up Ferrule can't place | Those NICs send nothing of their own and carry no VM, or are on a switch Ferrule doesn't read. Turn on LLDP on the host |
| The wrong port is shown as a host's | Tell us. A port should never be counted if it carries another host's or switch's addresses, or has another neighbour on LLDP |
| A free port isn't counted | It's reserved, disabled, an uplink, or the wrong kind (an SFP cage for a 10G link) |
| BMC port not known | No Redfish source for that host's BMC, or its BMC lists no network port. Ferrule matches BMC to host by serial number, then host name |
| Which UPS feeds the rack? | More than one UPS, and no rack says which. Set the rack's UPS on Settings |
| No PDU outlet is named for the host | Name its outlets on the PDU after the host (such as esx01 PSU1) |
| PDU: free outlets not known | The PDU doesn't meter each outlet, so free can't be told from in use |
| Rack space not known | Connect NetBox, or set the rack on Settings |
| An address line says no subnet | No subnet in the IPAM holds that address. Add it to the IPAM |
| vSphere licenses not read | Give vCenter's account the Licenses privilege |
Reclaim¶
| What you see | Why, and what to do |
|---|---|
| 0 of N VMs have the 21 days it takes to judge | Per-VM history starts from the first poll after this release. Three weeks |
| A busy VM is called idle | Its CPU never passed 5% of its vCPUs all month. If it's a VM that wakes once a month, ignore it |
| A VM isn't called oversized on memory though it should be | The guest doesn't report its own memory use (no VMware Tools). Ferrule doesn't guess |
| Orphaned disks: can't be browsed | Give vCenter's account Browse datastore. Each datastore it couldn't search is named |
| Orphaned disks: not accessible | The datastore is unmounted or offline, so it wasn't searched |
| A disk listed as orphaned is in use | It belongs to a vCenter or cluster Ferrule doesn't watch, on a shared datastore. Check before deleting |
The monthly summary¶
| What you see | Why, and what to do |
|---|---|
| Not emailed: set SMTP_HOST and SMTP_FROM | The summary uses the mail server alert email uses. Set it up |
| Not emailed: no address | Set Send to on Settings, or CAPACITY_SUMMARY_TO |
| The last attempt failed | The error is shown. It retries hourly through the 1st |
| It didn't arrive on the 1st | It's sent from 07:00 local time: check TZ. And that the license was in force then |
| "What to buy" is empty | No cluster is within six months of its threshold |
Questions¶
Do I need a database? No. Everything is kept in files on Ferrule's volume. The reporting database is optional, for Power BI or Grafana.
Does it slow my vCenter or Proxmox down? No more than before: capacity planning reads what Ferrule already polls. The extra reads, licenses (hourly), datastore browsing and storage content (once a day, with the inventory), and switch MAC tables (every ten minutes), are light.
Can it see what I'm about to deploy? No: forecasts continue the past. Use What if for planned projects.
Why is a host's CPU shown but never alerted? A cluster moves VMs between hosts. A busy host is a balancing question, not a capacity one.
What happens if the license lapses? The tab, forecasts, alerts, metrics and summary stop, and reserved ports stop being shown. Ferrule keeps collecting the daily history, and keeps your settings, reservations and racks, so nothing is lost when a license is added again.
Is anything written to my hypervisors, NAS, IPAM or DCIM? Never. Every source is read-only, and every suggestion is for you to act on.
What about MSPs with many sites? One summary per site comes with the hub for remote collectors. Until then, each install plans and emails its own.