Australian cloud for business software — built and run in Sydney since 2002

NetLeverage
NetLeverage

File drops and SFTP feeds into hosted business applications

Technical article
Hosting

About this article: This article is reviewed against the cited sources. This is source research, not lab validation. Check the stated versions and your own requirements before making changes.

Direct answer: Some hosted business applications still depend on batch files—invoices, price lists, payroll extracts, lab results—delivered by SFTP, folder drop, or a scheduled copy into a watched directory. Reliability comes from agreeing format, timing, idempotency, failure alerts and ownership across the sender, the hosting environment and the application, not from picking SFTP instead of email attachments. Treat the feed as a small service with an SLA, even when no public API exists.

Why it matters

File-based integration is sometimes chosen because the ISV or legacy module exposes import only from disk, because a partner cannot call HTTPS APIs, or because compliance prefers a clear “file landed here at time T” audit trail. When the pattern moves onto hosted servers, the fragile parts are usually network path, identity, clock, disk space, duplicate processing and silent partial failures—not the SSH cipher suite.

Integrators who design only the happy path discover that a missed 02:00 drop still “succeeds” at the OS layer while the application never imports, or that a re-sent file creates double charges because the import job is not idempotent.

This article is steady-state hosted LOB file feeds. It complements—but does not replace—published guides on customer integrations (VPN, API and file drop) (ISV product angle), verifying backup delivery to S3 or SFTP, and SFTP host key changes.

How it works

Delivery models (generic patterns; exact implementation varies by hoster and application):

Pattern What moves Typical trigger What the application needs
Inbound SFTP Partner pushes to a server endpoint Cron on sender side Staging folder, virus/size checks, import job
Managed drop zone Customer or partner writes to agreed path Watch folder or schedule Permissions separated from production DB
Outbound pull Hosted job fetches from partner Scheduler on hosted side Firewall allow list, credential rotation

SFTP and SSH (sourced at high level): SFTP is a file-transfer protocol carried inside an SSH session; the file transfer format is specified in the IETF document draft-ietf-secsh-filexfer-02 (internet draft, not an RFC). RFC 4253 describes the SSH transport layer that underpins the encrypted session; it does not define SFTP itself. OpenSSH sftp-server(8) documents how a common SFTP server implementation exposes the subsystem.

Operational mechanics (practitioner guidance—not vendor specification): hosted LOB teams should document for each feed:

  1. File contract — encoding, delimiter, naming convention, maximum size, required trailer record, timezone on timestamps.
  2. Window — earliest/latest arrival; what “late” means for the business day.
  3. Staging — land in incoming, then promote into ready only after the upload is fully closed and validated against the partner’s expected byte size, hash or manifest—not a size-settle timer alone. A rename (for example .part to the final name) exposes a completed, validated file for watchers; it does not by itself prove completeness. Rename can replace the destination name atomically only when source and destination are on the same filesystem and the server supports atomic rename semantics—do not equate “same SFTP server” or “same path” with same filesystem. Confirm behaviour in your environment. On mismatch, quarantine, alert and do not import.
  4. Processing — a single-worker lock reduces concurrent imports on one host, but batch IDs alone do not deduplicate across crash and retry. Design durable idempotency: transaction boundaries, duplicate detection, or application-level reject of repeated batches; move successes to done or quarantine error with a reason code.
  5. Notification — success and failure visible to both customer ops and the hoster support path; agree whether alerts are portal-visible or ticket-driven.

Partial files are a recurring defect: partners upload slowly while the import job starts at byte zero. Hash-or-manifest checks before promotion, with rename used only as a visibility step after validation (where atomic rename semantics are confirmed), reduce that class of incident better than waiting for a stable file size alone.

What to choose and why

(Practitioner guidance—not vendor specification.)

SFTP vs shared folder vs API: choose file drops when the application or partner capability is fixed on files. Prefer key-based SFTP with IP allow lists when the partner is stable; avoid exposing broad internet-facing drop endpoints without rate limits and monitoring. If the business later needs near-real-time events, plan a parallel channel rather than polling the same folder every minute—polling masks errors and loads disks.

Hosted placement: agree whether the drop zone lives on an application server, a dedicated integration VM, or customer-controlled storage the hosted job reads. Each choice shifts backup, AV scanning and patch responsibility under your shared responsibility model.

Idempotency: design imports so a repeated file either replaces the same logical batch (versioned batch ID) or is rejected with a clear log line. Finance and inventory modules punish duplicate imports severely.

Risks and limits

No SSH server configuration or import scripts were executed here (untested). Virus scanning, DLP and Australian privacy obligations for file content remain customer-and-contract specific.

Do not assume any hosting provider offers a universal “integration SFTP SKU”; negotiate endpoints, credentials custody and log retention in service discussions.

Sources

Learning notes

  1. A successful SFTP login does not prove the business import ran.
  2. Completeness needs hash, size or manifest confirmation—not size-settle alone.
  3. Each feed needs an owner on both customer and hosting sides.
  4. Idempotency must survive retries, not only a single-run worker lock.
  5. Hosted placement of the drop zone determines who patches and backs up the path.