createSessionWithTimeout checked sys.client for nil and then dereferenced
it again inside the goroutine that calls NewSession. update() runs the
SMART fetch in its own goroutine, so closeSSHConnection can clear the
field between those two reads and the goroutine dereferences a nil
client, panicking the whole hub process.
Make client an atomic.Pointer, load it once before starting the
goroutine, and clear it with Swap so a concurrent close cannot be
observed mid-session-creation. NewSession on an already-closed client
returns an error, which the existing retry path already handles.
Closes#2157
- Report battery data for individual devices
- Select a representative battery for legacy fields and alerts
- Average named battery data independently
- Display multiple batteries in system charts
- Add cross-platform coverage and transport tests
intel_gpu_top does not support the xe driver, so skip it for xe devices
and let the existing nvtop last-resort collector handle them. nvtop leaves
device_name unset on xe, so name the GPU from its PCI device id ("Intel GPU
(<id>)"). Adds nvtop to the Intel agent image (gputop already ships with
igt-gpu-tools).
The per-system updater reused an SSH client across ticks and ran the data
exchange with no deadline. If a connection went half-open (dead peer that
never sends RST/FIN) or an agent accepted the session but never wrote a
response, the read in fetchDataViaSSH blocked forever. Because
StartUpdater calls update() synchronously on its ticker, a blocked read
froze the whole per-system goroutine: the ticker's subsequent ticks were
dropped, no error was returned so the system stayed "up", and the agent
was never re-dialed until the hub process restarted.
Bound each SSH data exchange with sshOperationTimeout via runWithTimeout:
on timeout the connection is torn down (unwinding the blocked read) and a
retryable error is returned, so the next tick re-dials. Also enable TCP
keep-alive on dialed connections as a backstop for genuine network death.
Fixes#2041
CalculateCpuPercentLinux subtracted the stored previous counters from the
current ones without checking direction. When a stats response is processed
after a newer one for the same container, or an accounting counter resets,
the current total reads lower and the unsigned subtraction wraps to ~2^64
instead of going negative.
On the container counter that surfaces as the reported error, and the sample
is discarded along with the container's network stats:
cpu pct greater than 100: 1.15292150348562e+13
On the system counter it is quieter and worse: the wrapped value inflates the
divisor, so the percentage collapses toward zero and is stored as a healthy
sample rather than rejected. A synthetic rollback measures 2.7e-12 percent.
Both directions are now treated as a new baseline (0% for one sample), which
matches how the function already handles the first-run case.
CalculateCpuPercentWindows had the same unguarded subtraction and is fixed
the same way.
Fixes#2149
Co-authored-by: Ryan Chou <ryanchou1994@users.noreply.github.com>
Switch to localStorage, which is shared across all tabs/windows of the
same origin. Also clean up the stored entry and strip the callback URL
on both success and failure paths.
valibot's email() action uses a domain-label regex that disallows the
-- sequence, so Punycode ACE labels like xn--mnchen-3ya.de (the ASCII
form of münchen.de) are incorrectly rejected.
Switching to rfcEmail() applies the RFC 5321 domain-label pattern,
which allows hyphens within labels and therefore accepts both standard
and internationalized domain names.
Some enterprise NVMe drives (e.g. Dell Ent NVMe CM7 U.2) report capacity
via nvme_total_capacity instead of user_capacity.bytes in smartctl output.
The NVMe SMART parser now falls back to nvme_total_capacity when
user_capacity.bytes is zero.
- move hub's GetEnv function to new utils package to more easily share
across different hub packages
- change System.HasUser to take core.Record instead of user ID string
- add tests
