A pressure-hold visual inspection on the factory floor remains the baseline gate for shipping a liquid-cooled rack, with the Open Compute Project's liquid cooling integration and logistics guidance requiring confirmation that no gas has escaped from the secondary loop before coolant fill [S1].
Once the rack is in service, that pre-shipment test is reinforced by a permanent detection layer: leak sensors and leak-sensing wires that flag saturation in real time, plus flow and pressure monitoring on each coolant distribution unit (CDU) to catch slow drips and pressure drift before they escalate [S5][S6].
Why a Pressure-Hold Test Before the Coolant Goes In
The OCP liquid cooling integration and logistics white paper treats the factory leak test as a simple visual inspection, but its position in the workflow is not trivial: a quality check confirms gas has not leaked from the system and is required prior to fill, with visible signs of a pressure-side leak screened out before any water or glycol mix is introduced [S1].
Shipping a sealed loop charged only with shop air (or nitrogen) at a controlled test pressure lets the assembler find joint and seal defects without contaminating the secondary loop, and it keeps the leak-localization problem tractable, because the alternative, finding a leak after dielectric fluid or glycol is in the circuit, turns a 5-minute joint re-torque into a full drain, flush, dry, and refill job on a 30 kW-class rack [S3][S4].
Two-Loop Architecture Sets the Leak-Containment Rules
Liquid-cooled data halls almost universally run a primary loop on the facility side and a secondary loop on the IT side, joined through a heat exchanger; the primary loop is typically a 20 to 40 percent propylene glycol and water mix, while the secondary loop feeding cold plates and rear-door heat exchangers leans on lower glycol concentration or pure water to preserve heat-transfer performance [S3].
The pressure split matters for leak testing on site: the primary loop runs at the higher pressure needed to move fluid up to cooling towers and across the data hall, while the secondary loop is held at a lower, IT-friendly pressure, so a leak inside a server rack vents inward into the room rather than depressurizing the building's chilled water plant [S3]. A 30 percent propylene glycol mix freezes near minus 12 degrees C, and operators are increasingly pairing that fluid chemistry with pH and resistivity checks because corrosion inside the loop, not just a bad fitting, is one of the two root causes called out for coolant leakage [S3][S5].
Detection Stack That Backs Up the Factory Test
Eaton's published approach layers leak sensors and leak-sensing wires at the rack, with the wires detecting saturation and promptly indicating the presence of a leak so the building management system can isolate the affected rack before a slow drip becomes a short [S5]. The same vendor pairs barb fittings with tube clamps, solders or brazes metal joints, and installs in-line filters inside the CDU to keep larger particles from cutting O-ring seals, a sequence that has to pass the pre-shipment leak test before the rack is crated [S5].
Texas Instruments' reference design on leak detection, flow, and pressure monitoring frames the on-site layer as an early-alert system: sensors warn of leaks or other issues before physical damage occurs to the server racks, which is functionally the production counterpart of the OCP visual check [S6]. For glycol-water mixtures, Dober's monitoring guide adds a coolant-chemistry tier on top of the leak-detection tier, tracking pH drift, pressure decay, and resistivity so that a coolant loop that passes the visual test on day one is still flagged when its inhibitor package degrades over months of service [S4].
Failure Modes the Pre-Shipment Test Has to Catch
Corrosion and fluid interconnect issues are the two root causes of coolant leakage cited for liquid cooling systems, and they are not equally visible during a factory pressure-hold: a bad QD or solder joint shows up immediately as a bubble or pressure decay, while galvanic or erosion corrosion is internal and often surfaces only after months of operation, which is why the on-site detection layer is designed to pick up what the visual inspection cannot [S5].
Quick-disconnect fittings and swivel joints are the specific interconnect types that undergo accelerated testing in the factory, with QD targets set for dripless disconnects and hot-swappable service so that the same joint that passes the pre-shipment leak test can be cycled in the field without re-pressurizing the entire loop [S5]. The cost of missing either class of defect is reflected in the downtime numbers cited across the data center industry: 60 percent of outages in 2022 cost more than 100,000 US dollars, up from 39 percent in 2019, and the share exceeding 1 million US dollars rose from 11 percent to 15 percent over the same window, with water and coolant events persisting in the top five root causes of unplanned outages [S4].
Comparison: Pre-Shipment Test vs On-Site Detection Layer
Four criteria separate the factory test from the production monitoring stack, and the two are designed to complement each other rather than substitute for each other [S1][S4][S5][S6].
Scope: the pre-shipment pressure-hold covers the secondary loop from CDU outlet to rack inlet and back, with the loop charged with gas and inspected for visible leakage; the on-site layer covers the same loop plus the facility primary loop, with continuous sensors, flow meters, and chemistry probes.
Leak type caught: a factory test reliably catches gross joint and seal failures (bad braze, misseated O-ring, loose QD clamp); the on-site layer is what catches slow drips, internal corrosion, and the slow pressure drift that the visual test would have missed [S1][S5].
Test medium: factory test uses shop air or nitrogen at a controlled test pressure so a leak is visible and the loop is dry; on-site detection runs against the actual coolant, typically a 20 to 40 percent propylene glycol and water mix, and the sensor technology has to be specified for that fluid [S3].
Response time: a factory failure is caught before crating, so the cost is a re-work; an on-site leak that escapes the sensor layer can take down a server, corrode PCB traces over months, and trigger a short-circuit event within seconds of contact with live hardware [S3][S4].
What the Standards and Vendor Docs Actually Say
The OCP liquid cooling integration and logistics white paper (revision 1.0.1) defines the baseline: a visual inspection to confirm gas has not leaked from the system, required prior to fill, with signs of a gas leak screened at the factory [S1].
Eaton's leak-prevention write-up adds the manufacturing and packaging controls that make the visual test meaningful: soldered or brazed joints, barb fittings clamped to tube, filters on the CDU to keep particulates off O-rings, and accelerated testing of swivel and QD fittings to verify dripless hot-swap behaviour [S5].
Dober's monitoring guide (May 2026) and the IoT-focused industry write-ups concur on the operating rationale: average rack power density has roughly doubled from around 6 kW in 2016 to about 12 kW in 2024, and ultra-dense AI/HPC racks can exceed 30 kW each, so the consequence of a leak in a modern hall is a server-temperature spike measured in seconds, not minutes [S3][S4].
Spec Boundaries Buyers Should Pin Down
Three numbers belong on the factory acceptance sheet for a liquid-cooled rack: the test pressure used on the secondary loop during the pressure-hold, the minimum dwell time at that pressure, and the pressure-decay threshold that counts as a pass, all of which are set by the rack integrator and should be disclosed on the shipping documentation alongside the OCP visual-inspection sign-off [S1].
On the on-site side, the same documentation should list the leak-sensor coverage map (which zones of the rack have a sensing wire, which have a point sensor), the coolant chemistry limits, including pH window and glycol concentration tolerance, and the CDU's flow and pressure transmitter ranges, since pressure transmitter selection and flow meter sizing drive whether a slow drip is even visible to the building management system. Operators are increasingly pairing this instrumentation with point and cable leak detection on the floor and under the rack, in line with ASHRAE and OCP monitoring guidance, so the rack ships with a documented detection envelope rather than just a passing factory test [S4][S6][S7]. For facilities that also store spare capacity on pallet rack and storage rack systems near the data hall, the same leak-detection philosophy applies: a slow drip is a tracked signal long before it is a structural problem, and the rack integrator's documentation should make that signal visible at the BMS layer from day one.
Trackable signals going forward: a revision update to the OCP liquid cooling integration and logistics white paper beyond 1.0.1, and any vendor disclosure of a standardized test-pressure and dwell-time specification for the pre-shipment pressure-hold, both of which would convert the current visual-inspection baseline into a numeric acceptance gate.
For related coverage, see Mechanical Dock Leveler Spring Tension: Adjustment Procedure and Field Limits.