> For the complete documentation index, see [llms.txt](https://docs.pharos.ai/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.pharos.ai/open-telemetry/hermes/argus/elf.md).

# EventLogFile

ELF pulls after-the-fact EventLogFile records over REST — the only horizontally scalable Argus workload, with sharded workers and a live-vs-backfill queue.

ELF (`argus-elf`) pulls after-the-fact bulk CSV log files — Salesforce `EventLogFile` records — over REST. Unlike RTEM, which is capped by Salesforce's single-active-subscriber limit, ELF is the **only horizontally-scalable Argus workload**: you can run more workers to pull more log files in parallel.

***

## Sharding

The unit of work is one `EventLogFile`, claimed by record Id through a lightweight **lease** — a conditional write plus a TTL, with no heartbeat required. A worker that holds the lease downloads and parses that file; once the lease expires unclaimed, another worker can pick it up. The lease store is selected with `LEASE_STORE` (see [`ORG_CONFIGS` Reference](https://docs.pharos.ai/open-telemetry/configuration/org-configs.md#state-store-selectors)); a single replica per org needs none.

Each file is downloaded as a stream and parsed **row by row**, so memory use stays bounded by worker parallelism and batch size rather than by file size — a multi-gigabyte `EventLogFile` doesn't need to fit in memory at once.

***

## Discovery Loops

ELF discovers new `EventLogFile` records through two independent polling loops per event type:

* **Daily** — the default, authoritative loop. It's unfiltered and lands hours to a day after the underlying events occurred.
* **Hourly** (`Interval='Hourly'`) — lands within the hour, but requires the Salesforce Event Monitoring hourly add-on and a configured dedup store; without both, the Hourly loop is a silent no-op rather than an error.

Operators tight on Salesforce API budget can disable the Hourly loop with `hermes.elf.disable_hourly`.

***

## Live vs. Backfill

ELF runs live and backfill work through a **bounded priority queue** that always prioritizes live processing over backfill. When the pipeline can't keep up, ELF applies backpressure by **holding the cursor and re-fetching later** rather than spooling events locally — Salesforce's own **30-day `EventLogFile` retention** is the durable buffer, so there's no separate spool to manage or lose.

Backfill is a **one-off manual** operation, started with `-mode=backfill` and scoped by `hermes.elf.backfill` (`"full-window"` or `"from:YYYY-MM-DD"`). It runs in its own **isolated cursor namespace**, separate from live processing, so a backfill run can never interfere with the live cursor. Backfill jobs are **idempotent and resumable**: rerunning with the same `job_id` picks up where the previous run left off rather than reprocessing from scratch.

***

## Coverage

**34 EventLogFile event types have dedicated mappings** (Salesforce exposes roughly 70 event types in total). Adding a type is a mapping-config change — see [Pipeline Config](https://docs.pharos.ai/open-telemetry/configuration/pipeline-config.md). `InsufficientAccess` is an example of a log-only security type: a record-access denial with no RTEM equivalent.

***

## Scaling Signal

ELF's autoscaling is driven by the `elf_lag_seconds` gauge — how far behind the live cursor is from the present — published to the cloud autoscaler as `argus_elf_lag_seconds`. Size worker counts from observed lag in your own deployment, using the load-unit model in [Sizing & Cost](https://docs.pharos.ai/open-telemetry/hermes/sizing-and-cost.md).

Per-worker decode throughput (decode → mapping → identity → dedup, no export) was measured with a Go benchmark on an Apple M1 Max workstation on 2026-05-30 at about 64 MB/s; the Graviton (≈24 MB/s) and x86 Fargate (≈13 MB/s) planning figures are derived from per-core ratios of 1/2.7 and 1/5. See [Scaling Profiles](https://docs.pharos.ai/open-telemetry/hermes/scaling.md) for the `scaled`-tier worker-count rule of thumb.

Representative ELF metric names: `argus.elf.apex.cpu_time`, `argus.elf.api.call.count`.

![argus\_\* metrics emitted from Salesforce Event Monitoring, explored in Grafana.](https://4264222071-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2Fdl9tmJ56lvwg5XY63QYr%2Fuploads%2Fgit-blob-3b0e7e4576ed274b124d975cb130b40b137d04c4%2Fgrafana-metrics-argus.png?alt=media)

***

## Related

<table data-view="cards"><thead><tr><th></th><th></th><th data-hidden data-card-target data-type="content-ref"></th></tr></thead><tbody><tr><td><strong>👁️ Argus</strong></td><td>The Salesforce Event Monitoring bridge</td><td><a href="https://docs.pharos.ai/open-telemetry/hermes/argus.md">Argus</a></td></tr><tr><td><strong>⚡ Real-Time Event Monitoring</strong></td><td>Platform Events over the Pub/Sub API — the other Argus surface</td><td><a href="https://docs.pharos.ai/open-telemetry/hermes/argus/rtem.md">Real-Time Event Monitoring</a></td></tr><tr><td><strong>📐 Sizing &#x26; Cost</strong></td><td>Load units, worker counts, and cloud translation factors for ELF</td><td><a href="https://docs.pharos.ai/open-telemetry/hermes/sizing-and-cost.md">Sizing &amp; Cost</a></td></tr></tbody></table>
