Elglide vs. Snowflake Openflow
Snowflake Openflow and Elglide both land your operational databases in Snowflake. The difference shows up before the first sync, in what you have to stand up and what it costs while idle, and on the sources Openflow’s database connectors can’t reach: databases where log-based change capture was never an option, and views, which never appear in a write-ahead log at all.
Nothing to stand up
Register, enter a host and a password, test, pick tables, pick a schedule. No deployment, no runtime, no roles to grant, no driver to upload. Inside your network, one install package runs the agent.
When CDC isn’t an option
Log-based capture needs privileges and source-side changes
plenty of databases will never get. Elglide keeps syncing
incrementally with nothing but SELECT, and syncs
views the way it syncs tables.
Rows, not runtime-hours
$49 per 1M rows synced, the first 1M every month free, and the initial full load never billed — one line on the bill, metered live in-app. No always-on runtime to fund: an idle Openflow deployment bills about 80 credits a month before a runtime even starts.
Elglide vs. Snowflake Openflow
| Features | Elglide | Snowflake Openflow |
|---|---|---|
| Getting set up | ||
| Before the first sync | ✓Register (no card), add a connection — host, port, user, password — test it, pick tables and views, pick a schedule, run. Source inside your network? Download one install package from the Agents page and run the installer. No registry login, no cloud account, no inbound port. |
A Snowflake administrator creates an Openflow admin role
and grants it account-level privileges, then a
management database, an execute-as role, a network rule
and an external access integration for each source
host, a secret for the source password, a deployment,
and a runtime whose size you choose up front. You
download the JDBC driver from the database vendor and
upload it in the setup wizard. Role and runtime creation
repeat for every runtime.
Gen 2 (generally available
September 2026) made deployments and runtimes SQL
objects and added a setup wizard; the list above is the
gen 2 path, and the wizard itself is still in
Public Preview. Users whose default role is
ACCOUNTADMIN cannot log in to a
runtime.
|
| Changing your mind later | ✓Schedule, table list and sync mode change in place from the panel. Nothing is torn down or recreated. | The runtime’s node type is fixed at creation; a different size means a new runtime and either a full re-snapshot or a connector migration. Destination schema pattern, identifier resolution and table storage format cannot be changed once ingestion starts. |
| Getting to incremental | ||
| Source-side setup required |
✓RowHash and
Watermark need only SELECT. No schema changes,
no replication privileges, nothing created on your source
database.
|
PostgreSQL needs wal_level = logical (an
ALTER SYSTEM and an instance restart), a user
with the REPLICATION attribute, a publication,
and REPLICA IDENTITY configured per table. SQL
Server needs change tracking or CDC enabled; Oracle needs
XStream.
|
| When CDC can’t be enabled | ✓The table drops down the ladder — Watermark, RowHash, or Full — and keeps syncing. The cases that actually come up: a vendor-managed application database where a config change voids support, a managed instance where logical replication or the binlog was never switched on, no privilege to grant, or a DBA who won’t accept the WAL overhead. | ✗The database connectors are built on native change capture. Meet the prerequisites, or move that source to another tool. |
| Oracle licensing | ✓LogMiner. No additional Oracle license needed. | XStream. Oracle’s own documentation, quoted by Snowflake, makes Enterprise Edition “a prerequisite to license and use Oracle XStream” — so you take an embedded license through Snowflake or bring your own GoldenGate entitlement. Snowflake’s published price for the embedded license: $70 license plus $40 support and maintenance per licensed core per month, on a non-cancelable 36-month term (a 12-month term exists for public sector). Opt out of support later and the connector locks. |
| Tables without a primary key | ✓Full mode always works and needs no key at all. A key matters only when you want to climb to an incremental mode. | The PostgreSQL connector “requires every replicated table to have a supported identity key configuration.” The SQL Server connector “only replicates database tables that contain primary keys.” |
| Delete detection | ✓RowHash catches physical deletes with no log access — a row missing from this run’s read is deleted in Snowflake on the same merge. Watermark catches soft deletes (a status flag set by an UPDATE). Physical deletes leave a watermark nothing to find, so a per-run count audit demotes the table back to RowHash automatically if they appear. | Deletes arrive on the change stream — which means every prerequisite above has to be satisfied first. |
| What counts as a source | ||
| Database views | ✓First-class. Full sync always; incremental via RowHash once a unique key is confirmed against the destination. | ✗A view writes nothing of its own to the WAL or redo log. Every database connector page describes table replication only. |
| Source coverage | Databases — SQL Server, PostgreSQL, MySQL/MariaDB, Oracle — plus files (CSV, TXT, JSON, XML, Parquet, Excel) from SFTP, Amazon S3, Windows or NAS shares and folders on your servers, and REST APIs. | ✓Much broader. Kafka and Kinesis, Salesforce, Workday, HubSpot, Shopify, Jira, Dataverse and Veeva, plus unstructured content from Box, SharePoint, Google Drive and Slack. If your sources live there, Openflow covers ground we don’t. |
| Structured files on a file server or SFTP |
✓Ready-made. Point a
connection at \\server\share or an SFTP server,
pick the files by pattern, and each set loads into its own
table. An agent inside your network reads them in place
— no staging copy in the cloud.
|
✗No ready-made connector. Openflow’s file connectors (Box, SharePoint, Google Drive) load unstructured documents for Cortex; loading CSV or Excel from a share or SFTP means building a NiFi flow from processors yourself. |
| Deployment & freshness | ||
| Runs inside your network | ✓The hybrid agent runs on your side of the firewall and opens one outbound mutual-TLS connection — no inbound ports. Data goes straight from your database to your Snowflake; the control plane stays in our cloud. | ✓BYOC runs the data plane in your own VPC (generally available in AWS commercial regions), or Snowflake hosts it for you on Snowpark Container Services. Real parity. Both designs keep your data on your side; pick on operational fit, not on this row. |
| Schedule granularity | ✓Per table and per view. Each object carries its own cadence — a 1-minute schedule on the table that matters, nightly on the one that doesn’t, on-demand on the rest — with a pipeline-level default for everything you don’t override. | Connector-wide. “Merge Task Schedule CRON” is a single parameter per connector instance, and the tables it covers are chosen by an include list or regex, so every table in that connector shares one cadence. Different cadences mean separate connector instances to configure and pay for. |
| Freshness | Scheduled runs, not a live stream. Freshness is whatever cadence you set, down to a minute; nothing runs in between. Elglide does not do true streaming today — that is a later phase. If you need sub-second latency, say so and we will tell you straight. | ✓Continuous near-real-time capture, with the option to throttle destination merges on a CRON schedule. Openflow is ahead here. |
| Throughput |
Each run is one Parquet batch per table — extract,
one PUT, one COPY or
MERGE — so your warehouse works for
seconds per table per run and nothing runs in between.
We publish no events-per-second
figure. For a sustained high-volume change stream,
Openflow’s continuous capture is the better
fit.
|
✓Sized by sustained events per second: a Small runtime below about 1,000, a Medium for 1,000 to 5,000 hosting 5 to 8 connectors, a dedicated runtime for a source above about 15,000. Snowflake calls these starting points to measure and adjust, and the node type is fixed at creation. |
| Cost | ||
| What the bill is made of | ✓One line: $49 per 1M rows synced, the first 1M every month free, and the initial full load never billed — metered live in-app (see pricing). A database changing 5M rows a month costs $196. Nothing is billed while nothing syncs. | Several lines, at Snowflake’s published rates: runtime compute at 0.11 (Small), 0.41 (Medium) or 0.83 (Large) credits per hour, billed per second with a 5-minute minimum on a Snowflake-hosted deployment, or 0.0225 credits per vCPU-hour plus your own EC2, EKS and storage bill on BYOC; warehouse credits for the initial snapshot and every CDC merge; Snowpipe ingest; telemetry per GB. The management pool “continues to run and incurs costs, even if there are no runtimes running” — one CPU_X64_S instance per deployment. A deployment plus one always-on Small runtime comes to about 160 credits a month: $320 at the $2 Standard rate, $480 at the $3 Enterprise rate (AWS US East, on demand), before the warehouse runs a single merge. |
| Where Openflow wins | ||
| Native to Snowflake | A separate vendor, a separate contract, a separate bill. | ✓It is a Snowflake product: billed on your Snowflake contract, managed from Snowsight, supported by the same account team. For a Snowflake-first shop it is the incumbent default, and that counts for a lot. |
| Beyond databases | Databases, files and REST APIs. That is the whole product, and it is where our engineering goes. | ✓Built on Apache NiFi with hundreds of processors, covering streaming, SaaS applications and unstructured content — structured and unstructured text, images, audio, video and sensor data. |
Comparison based on Snowflake’s published Openflow documentation and the Snowflake Service Consumption Table effective October 15, 2026, as read in October 2026 (connector list re-checked October 24, 2026); Openflow ships quickly and capabilities and rates may have changed since. Credit prices are Snowflake’s on-demand list for AWS US East; contracted rates differ. Sources: About Openflow, Snowflake deployments, setup tasks, gen 2 release note, Openflow connectors, PostgreSQL connector setup, SQL Server connector, Oracle connector, CDC runtime sizing, BYOC cost, Snowflake-deployment cost and the Service Consumption Table. Snowflake, Openflow, Snowsight, Oracle, GoldenGate, Apache NiFi and all other marks named here are trademarks of their respective owners and are not affiliated with Elglide.