feat(ansible): add net.core sysctl tuning params to tuned profile

[hunleyd]

What

Adds the following net.core.* parameters to the sysctl section of the PostgreSQL tuned profile (ansible/tasks/setup-tuned.yml):

option	value
net.core.netdev_budget	1024
net.core.netdev_max_backlog	10000
net.core.rmem_default	262144
net.core.rmem_max	104857600
net.core.somaxconn	16384
net.core.wmem_default	262144
net.core.wmem_max	104857600

It also adds a tracked .yamllint config (and a .gitignore negation so it survives the repo's .* dotfile ignore) so that standalone yamllint agrees with the yamllint profile ansible-lint already applies internally.

Why / Benefit to PostgreSQL

These kernel network-stack settings improve how Postgres handles connections and high-throughput client traffic:

• net.core.somaxconn (16384) — raises the ceiling on the listen backlog so bursts of incoming connections aren't dropped during connection storms (e.g. poolers reconnecting, many clients starting at once). Pairs with Postgres's own listen backlog.
• net.core.netdev_max_backlog (10000) — allows more incoming packets to queue when the NIC receives faster than the kernel can process, reducing packet drops under heavy network load and the retransmits/latency spikes that follow.
• net.core.netdev_budget (1024) — lets the kernel process more packets per softirq poll cycle, improving network throughput on busy database servers handling many concurrent connections.
• net.core.rmem_default / rmem_max (256 KiB / 100 MiB) — larger receive socket buffers let Postgres backends absorb large inbound payloads (bulk INSERT/COPY, large query parameters) without stalling on slow/high-latency links.
• net.core.wmem_default / wmem_max (256 KiB / 100 MiB) — larger send socket buffers let backends ship large result sets (big SELECTs, COPY TO, replication streams) without blocking, improving throughput over high-bandwidth-delay-product networks.

Together these raise the buffer and queue headroom of the network stack so a high-connection, high-throughput Postgres instance is less likely to drop connections or bottleneck on the kernel under load.

Notes

• net.core.somaxconn was previously set to 16834 — corrected to 16384 (transposed-digit typo).
• The .yamllint config mirrors ansible-lint's embedded yamllint profile (line-length off, document-start disabled, braces/truthy relaxed for Ansible). With it in place, setup-tuned.yml passes both ansible-lint and yamllint with no playbook content changes. Pre-existing lint debt in other, untouched task files is intentionally left out of scope.

INDATA-378

[hunleyd]

@claude review

[jfreeland]

this seems awfully high for free tier? cc: @danielmitterdorfer for review from #team-perf-eng

[hunleyd]

you're probably not wrong, but as I said on Slack, i'm just consolidating the values from elsewhere in our repos into tuned. once it's all in tuned, then we can talk instance sizing 🤞🏻