Fix Telegraf options removed in 1.40 on Linux (fieldpass -> fieldinclude, pid_tag -> tag_with)

[suyadav1]

Problem

The Linux agent ships Telegraf 1.39.x, which logs deprecation warnings for options Telegraf removes in 1.40.0:

W! DeprecationWarning: Option "fieldpass" of plugin "inputs.disk" deprecated since version 1.29.0 and will be removed in 1.40.0: use 'fieldinclude' instead
W! DeprecationWarning: Option "fieldpass" of plugin "inputs.diskio" ...
W! DeprecationWarning: Option "fieldpass" of plugin "inputs.net" ...
W! DeprecationWarning: Option "ignore_protocol_stats" of plugin "inputs.net" deprecated since version 1.37.0 ...
W! DeprecationWarning: Option "fieldpass" of plugin "inputs.prometheus" ... (x2)

These read as cosmetic, but they are not. Telegraf treats an unrecognized option as a fatal config-load error, not a warning:

E! loading config file failed: plugin inputs.disk: line 451:
   configuration specified the fields ["fieldpass"], but they were not used

The process exits with RC=1. An empty value (fieldpass = []) fails identically, so this is not limited to clusters that customize prometheus_data_collection_settings — a default ConfigMap hits it too. On a 1.40 bump, every Linux ama-logs Telegraf process would refuse to start, dropping InsightsMetrics and breaking the disk-usage MDM alert path (container.azm.ms/disk → used_percent → MdmMetricsGenerator.rb).

Fix

Migrate the Linux configs to the replacement option names:

Removed in 1.40	Replacement
fieldpass	fieldinclude
fielddrop	fieldexclude
pid_tag = true	tag_with = ["pid"]

Covers the daemonset, replicaset, Prometheus sidecar, and the separate ama-logs process-metrics Telegraf instance, plus the Prometheus stanzas generated at runtime by tomlparser-prom-customconfig.rb and tomlparser-osm-config.rb.

pid_tag (inputs.procstat) is not in the warning list above because that config runs as a separate, unguarded Telegraf process whose output isn't in the pod log tail — it is removed in 1.40 all the same, and was caught during validation.

Also drops ignore_protocol_stats from inputs.net. Protocol-stat collection was removed in 1.37.0 and the option has been a no-op since, so behavior already matches the previous true. (This one is scheduled for removal in 1.45, not 1.40.)

Windows is deliberately unchanged

Windows is pinned to Telegraf 1.24.2 (kubernetes/windows/setup.ps1), which predates fieldinclude/fieldexclude/tag_with — all added in 1.29.0. Renaming there would break it immediately. The Windows configs keep the legacy names, and the comment at setup.ps1:44 now documents the split so the next upgrader doesn't "fix" the inconsistency.

build/common/installer/scripts/tomlparser-prom-customconfig.rb is the only file shared by both platforms, so it selects option names via is_windows?, mirroring the existing timeout / response_timeout branch directly above it.

Startup validation gate

The gate in kubernetes/linux/main.sh was:

/opt/telegraf --non-strict-env-handling --config <conf> --input-filter file -test

--input-filter file restricts loading to the dummy file input, so it returns RC=0 for a config that is fatally invalid — the broken config gets promoted, and the real (backgrounded, unchecked) Telegraf start then dies silently. I verified this against the 1.39.1 binary: the old gate passed a config that the agent could not actually load.

Replaced with telegraf config check, which validates every configured plugin and performs no gathering. Note that simply dropping --input-filter file from -test would have been a regression — -test really does scrape kubelet/cadvisor, so it could fail on unready endpoints and silently discard customer Prometheus config.

--non-strict-env-handling must come after the config check subcommand; in the global position it is silently ignored.

Validation

Run against the real Telegraf 1.39.1 binary in a live AKS cluster, using each config's fully substituted runtime form:

Config	Deprecation warnings	config check
telegraf.conf (daemonset)	6 → 0	RC=0
telegraf-ama-logs-process-metrics.conf	9 → 0	RC=0
telegraf-rs.conf (replicaset)	0 → 0	RC=0
telegraf-prom-side-car.conf	0 → 0	RC=0
Generated stanzas (namespaced prom + OSM)	0 → 0	RC=0

Also confirmed:

• inputs.net emits an identical field set before/after the rename — no metric drift.
• Upstream's own telegraf config migrate independently produces the same mapping for all five affected plugin types (disk, diskio, net, prometheus, procstat). Its output wasn't taken wholesale because it reorders and reformats the entire file; the renames here are surgical.
• Key ordering is preserved so plugin-level keys stay above nested tables ([inputs.disk.tagdrop], [inputs.procstat.tags]) — otherwise they'd bind to the nested table, which is what the existing influxdata/telegraf#5615 "ORDER matters" comment guards against.
• is_windows? branch verified functionally: emits fieldinclude/fieldexclude on Linux, fieldpass/fielddrop on Windows.
• ruby -c clean on both parsers; bash -n clean on main.sh.

Not changed

• Customer-facing ConfigMap keys — fieldpass/fielddrop under prometheus_data_collection_settings in container-azm-ms-agentconfig.yaml are a public API and stay as-is. The new names are used only when rendering Telegraf TOML.
• TELEMETRY_*_FIELDPASS_LENGTH env var names.
• Windows configs (see above).
• The 1.40 upgrade itself. This makes the codebase 1.40-ready; the version bump is a separate change.

[azure-pipelines]

Azure Pipelines:
There may be pipelines that require an authorized user to comment /azp run to run.

[suyadav1]

/azp run

[azure-pipelines]

Azure Pipelines:
Successfully started running 1 pipeline(s).
2 pipeline(s) were filtered out due to trigger conditions.