log-explorer Helm values reference

Helm values reference table for the log-explorer application.

Key	Type	Default	Description
affinity	object	{}	Affinity rules for the log-explorer deployment pod
cache	object	See values.yaml	Persistent volume for the log cache. Fetching a night out of Loki takes minutes, so the cache is what makes the tool usable a second time; an emptyDir would throw it away on every pod restart.
cache.mountPath	string	"/var/cache/ra-log-explorer"	Where the cache volume is mounted, and what the application is told to use as its cache root.
cache.sizeGi	int	50	Size of the cache volume, in GiB. Given as a bare number rather than a quantity string because it sets two things that must not disagree: the volume’s size, and the byte ceiling the application’s LRU eviction keeps the cache under (usagePercent of it). Deriving both from one value is what stops the app happily filling a disk it thinks is bigger than it is.
cache.storageClassName	string	"rook-ceph-block"	Storage class for the cache volume
cache.usagePercent	int	85	Percentage of the volume the application will fill before it starts evicting least-recently-used windows. The headroom covers filesystem overhead and the gap between a fetch finishing and the eviction sweep that follows it — the cache is briefly over its ceiling in between.
config.liveLagS	int	60	How far behind “now”, in seconds, each live poll stops. Covers Loki ingestion lag: a log line stamped t can be ingested seconds after t, and a poll that raced it would miss it. Raise this if the end-of-night verification pass reports pods needing refetches.
config.livePollS	int	300	Poll interval, in seconds, for live mode: the server continuously fetches the current night’s logs (all pods) into its cache, so exposure views during the night are served by slicing what is already on disk instead of querying Loki. Nightly data volume lands on the cache PVC (~9 GiB measured for a busy night, before LRU eviction reclaims it), so cache.sizeGi must keep room for a few nights. 0 disables live mode and returns the app to purely on-demand fetching. 300 keeps the Loki query load modest; an exposure becomes viewable at most ~5 minutes after its shutter+5min ideal.
config.lokiPassword	string	""	Loki password in the clear, for bringing an environment up before its Vault secret exists. Do not commit a real value: this repository is public, and anything set here also shows in plain text in kubectl describe pod and the Argo CD UI. Pass it as a Helm parameter override on the Argo CD Application instead, and clear it once the loki-password Vault secret is populated. Empty (the normal state) reads the password from that secret, and the pod starts either way — with neither, log fetches fail with a legible error and the rest of the app works.
config.lokiUsername	string	"omega"	Loki HTTP basic-auth username. The password comes from the loki-password Vault secret. Both are set server-side because the application takes no configuration from the browser at all: one process serves every user, so a credentials field would let whoever filled it in last break fetches for everybody else.
config.site	object	See the per-environment values files	The one site this deployment serves. A site pairs a Loki cluster (where the logs live) with the ConsDB endpoint that owns shutter-close truth for that cluster’s data; the two must match or a dataId resolves against the wrong observatory. Singular on purpose: an instance on manke is BTS and one on yagan is the summit, and because the same 13-digit dataId exists at both with different obs_end values, an instance that could be pointed at the other one would return answers that looked plausible rather than obviously wrong. Needs name, cluster, namespace, lokiAddr and consdbUrl; consdbTokenFile is omitted because an in-cluster ConsDB Service address never passes through Gafaelfawr and so takes no bearer token. Rendered into the sites.toml the application reads.
config.windowAfterS	int	300	Seconds after shutter close that a fetch window ends. As above, a starting value rather than a cap. Worth raising where the pipeline runs slower than the exposure cadence.
config.windowBeforeS	int	5	Seconds before shutter close that a fetch window starts. This is the starting value of an editable field, not a cap: widening a window to catch a neighbouring exposure is a normal investigative move.
config.workers	int	8	Parallel log-download workers per fetch. Each is a logcli process streaming one pod’s window, so this trades cluster load and pod CPU against how long a night-wide fetch takes.
global.host	string	Set by Argo CD	Host name for ingress
global.vaultSecretsPath	string	Set by Argo CD	Base path for Vault secrets
image.pullPolicy	string	"IfNotPresent"	Pull policy for the log-explorer image
image.repository	string	"ghcr.io/lsst-so/ra_log_explorer"	Image to use in the log-explorer deployment
image.tag	string	The appVersion of the chart	Tag of image to use
imagePullSecrets	list	See values.yaml	Image pull secrets. Needed only if the GHCR package is made private; a public package pulls without credentials.
ingress.annotations	object	{}	Additional annotations for the ingress rule
ingress.path	string	"/log-explorer"	Path the app is served under. Passed to the application as RA_LOG_EXPLORER_BASE_PATH as well as being the ingress path, so the URLs it serves and the URLs the ingress routes cannot drift apart.
nodeSelector	object	{}	Node selection rules for the log-explorer deployment pod
podAnnotations	object	{}	Annotations for the log-explorer deployment pod
replicaCount	int	1	Number of web deployment pods to start. Must stay at 1: the log cache is a ReadWriteOnce volume, and loaded exposures live in the serving process’s memory, so a second replica would answer with different state depending on which pod the request landed on.
resources	object	See values.yaml	Resource limits and requests for the log-explorer deployment pod
tolerations	list	[]	Tolerations for the log-explorer deployment pod