Liquid-cooling air lock or blockage? A low-flow diagnostic guide

Is low coolant flow an air lock or a restriction?

Separate a liquid-cooling air lock from a restriction by recording branch flow, local pressure loss and pump state together: sustained recovery after an approved gas-removal intervention supports trapped gas, while excess component pressure loss at matched flow supports restriction. Gas can cause unstable flow or interrupt circulation, while a loaded filter, partly closed valve or obstructed connector can produce persistent low flow. Restore service only after the affected branches meet their flow requirements and the loop carries the required heat load within its temperature and pressure limits.

Start by locating the loss of service: one server branch, one rack, or every branch served by the coolant distribution unit (CDU). Compare the condition with the last accepted operating point. A post-service timestamp raises the priority of trapped gas, incomplete quick-disconnect (QD) engagement and incorrect valve lineup, but does not identify which occurred.

An air lock is a gas pocket that obstructs or breaks a liquid circulation path. Trapped pockets, moving entrained bubbles and dissolved gas released by warming or pressure reduction behave differently; Caleffi’s idronics 2 (2007) explains these distinctions. Noncondensable gas also differs from coolant vapor formed at inadequate local absolute pressure. [S7]

This guide addresses post-service diagnosis in single-phase, positive-pressure technology cooling systems (TCS). Identify the separate facility water system (FWS) across the heat exchanger and follow dedicated gas-management instructions for intentionally negative-pressure equipment. Bürkert distinguishes those architectures. [S10]

When branch flow or IT temperature violates an approved limit, apply the site’s load-reduction or isolation sequence before diagnostic adjustments. Record the affected asset and its actual limit. Use initial liquid-loop commissioning for filling and balancing, and direct-to-chip cooling paths to identify the serviced circuit.

Fault differential: which cause fits the whole pattern?

Use the whole hydraulic pattern to choose the next test. Anvilfield’s differential combines the cited manuals and hydraulic guidance.

Name each pressure-tap pair: filter loss, branch driving pressure and pump rise span different boundaries; static fill pressure is a local gauge reading. [S3, S5]

One branch versus every branch

One affected branch directs the first checks to its valves, hose bends, connector engagement and high points. Compare a healthy parallel branch under the same header conditions: adequate header differential pressure with one low-flow path focuses the search downstream. [S4] [S6]

Low flow across all branches directs attention to common filters, pump performance, supply availability, makeup and control setpoints. Changed bypass position or demand also moves the operating point. Persistent low flow does not exclude a stationary gas pocket, and fluctuating flow does not establish gas. [S2] [S3] [S5]

What should you measure before changing anything?

Capture synchronized total and affected-branch flow, pump command and actual RPM, control mode, pump inlet/discharge pressures, filter inlet/outlet pressures and manifold pressures. Add coolant supply/return temperatures, fluid product and concentration, reservoir or fill status, vent activity and IT load. Save alarms before resetting them. [S2] [S4] [S5]

With the load protected, Anvilfield recommends one-second samples for ten minutes before and after an approved intervention, retaining raw minima, maxima and event timestamps. Match faster acquisition to the sensor response and controller update rate when the event is shorter than that interval; one-second logging cannot establish subsecond failover performance.

Verify sensor zero, range, units, pressure type and calibration; independently check a reported zero flow. Compare pressure taps at equal elevation and bore, or correct elevation and velocity head before calling the difference component friction loss, following Hydraulic Institute’s System Curves guidance, updated July 19, 2024. [S15] Use telemetry calibration and measurement uncertainty to document the instrument chain.

Preserve the control state

Constant-pressure control can mask resistance by increasing speed; constant-flow control can mask it by increasing pressure. Identify whether the controller regulates pump-outlet pressure, pump differential pressure or remote manifold differential pressure. Record the controlled tap pair. [S2] [S5]

Tag override expiry and automatic reversion; see Table E. Compare authorized fixed-speed intervals with valve and bypass positions recorded. At constant flow compare component loss; at constant differential pressure compare flow and speed. The CDU controls and duty acceptance guide covers the broader control sequence.

Resolve small component losses with an instrument suited to that differential. Subtracting two large gauge readings can leave an error comparable to the loss being diagnosed; record both instruments’ accuracy and zero checks, or use an approved differential instrument. At zero flow, a corrected pressure difference across an isolated gas pocket is not a liquid friction-loss measurement. Verify the measurement chain before assigning that difference to fouling.

Map the high points and their local pressures

Map every place gas can collect and distinguish an installed vent from a survey location. OCP’s Guidelines to Rack Manifold Requirements and Qualification, §3.3, calls for bleed or vent valves at the manifold top to remove trapped air and a drain at the bottom for sediment removal and servicing. [S6]

Table B uses selected elevations with a stopped pump, a connected water column, density rounded to 1,000 kg/m³ and gravity of 9.81 m/s². Anvilfield estimates local gauge pressure by subtracting elevation head from the datum pressure: p(z) = 100 − 9.81z kPa. Elevation changes local pressure; it does not add net pumping head around a filled closed loop. [S3] [S15]

At a selected datum of 50 kPa gauge, the same sealed column has approximately −9 kPa gauge at the overhead crest. Substitute the actual coolant density and pump-on pressure profile for site work. [S7]

Why a vent can admit air instead of releasing it

Anvilfield’s diagnostic rule is to measure local vent pressure before opening, during discharge and after closure: positive pressure at the CDU does not establish positive pressure at an elevated or pump-suction location. An open vent below atmospheric pressure can admit air. Keep the vent closed when its local pressure is outside the installed venting procedure’s permitted range; correct the pressure condition within the loop’s limits before venting. [S7]

Brief pressure or flow changes during opening can result from discharge and makeup. Judge recovery after the vent closes and conditions settle; liquid-only discharge proves clearance at that location and condition, not at every parallel hose arch. Recurrent air requires expansion-pressure and air-separation checks, plus ingress and makeup investigation. Raise fill pressure only within every connected component’s operating envelope. [S2] [S7] [S8]

Read pressure-flow trends without confusing cause and effect

Compare local pressure loss at matched flow, coolant temperature, concentration and component lineup. A falling flow can reduce measured differential pressure while resistance increases. Table C gives Anvilfield diagnostic predictions. [S3] [S4] [S15]

Log vent opening, closure, gas and makeup. Sustained recovery at comparable driving pressure supports gas involvement; persistent excess liquid-filled loss supports restriction. Exclude pressure, speed and bypass changes before assigning cause. [S2] [S3] [S7]

Normalize a loss only when the flow regime permits it

Scale corrected component loss by the square of the reference-flow ratio for comparable, liquid-filled, constant-coefficient conditions: Δp_ref = Δp_measured × (q_ref/q_measured)². Hydraulic Institute’s System Curves (updated July 19, 2024) supports this approximation. [S15]

Anvilfield estimates that a component losing 18 kPa at 90 L/min would lose 50 kPa at 150 L/min under a quadratic-loss approximation, showing why a modest measured differential pressure can conceal increased resistance when flow has fallen. [S15]

Use same-elevation, same-bore boundaries. Caleffi’s idronics 16 (2015), Formula 4-2, uses a 1.75 exponent for smooth pipe; Anvilfield estimates 44.01 kPa and a 2.20 ratio for these inputs using that exponent. Turbulence alone does not make the square law exact. [S3]

Use the manufacturer’s curve for laminar cold-plate channels, active regulators, non-Newtonian fluids or strongly changing viscosity. Do not normalize across changing gas fraction as though it were a measured fouling factor. See flow and pressure balancing for the relationship between branch resistance and distribution.

What if gas removal improves flow but leaves a restriction?

Table D’s intermediate state separates the two repairs. Anvilfield estimates 32 kPa at 120 L/min for the same restricted liquid-filled component that loses 18 kPa at 90 L/min; both normalize to 50 kPa at the reference flow. Gas discharge with sustained branch recovery supports a gas contribution, while unchanged normalized filter loss keeps the filter investigation open.

After isolated restriction service and the necessary de-airing, the selected final state returns to the clean curve. Record both interventions, their makeup quantities and the settled conditions separately. The calculation uses the stated constant-coefficient model; actual fault assignment requires the corresponding gas observation, component inspection and matched-condition trends.

Could this be cavitation, control hunting or a measurement error?

Vapor cavitation is a suction-pressure and vapor-pressure problem. Evaluate inlet absolute pressure against coolant vapor pressure and the installed pump’s required net positive suction head (NPSH) at actual flow, accounting for velocity and inlet losses. A positive gauge reading alone does not establish adequate NPSH. [S9]

Record temperature, inlet pressure and the installed pump curve. Gas also causes noise; the Grundfos allowance in Table E applies only to its named pump family.

Flow movement correlated with speed commands, setpoints or valve motion calls for a controls investigation. Bubbles can interfere with some flowmeters; the installed meter manual controls fluid compatibility, accuracy and straight-run requirements. Resolve installation and independent-reading discrepancies before accepting a sensor trace as proof. [S4] [S5]

Normal TCS flow with warm supply coolant redirects attention to FWS conditions, heat-exchanger performance and load. Approach temperature alone cannot diagnose blockage. Record temperatures on the correct sides of the heat exchanger before assigning the fault. [S5]

What purge limits apply to this equipment?

Apply the installed model’s complete procedure and identify its operating stage before selecting speed, pressure or duration. Table E separates reservoir-based ALX and pressurized CDU 121 procedures.

Koolance’s ALX English v1.1 filling procedure, printed p.9, requires unplugging the AC supply before adding more coolant. Its troubleshooting section permits model-specific priming adjustments and temporary tilting with the fill port closed; those instructions do not transfer to fixed racks or pressurized CDU assemblies. [S1]

Before opening the fluid boundary, identify the isolated zone, protected load, approved vent tool, discharge route, available coolant, residual pressure and energy-control procedure. Keep leak detection and isolation response effective and discharge away from electronics. Plan cooling continuity during isolated service before taking the affected branch out of operation. [S2] [S6]

Vertiv’s SL-80277 Rev A (September 2025) initial fill requires unit-offline commissioning; transfer or isolate affected IT load. [S2]

Stop and restore the pressure boundary

Stop the intervention when the model’s suction/fill limit, approved load envelope, coolant supply or discharge containment cannot be maintained. Do not defeat a protective trip to keep an IT load running during diagnosis. Commissioning overrides belong to the model’s approved unloaded sequence; filter opening and hose repair require the prescribed isolation and depressurization. [S2] [S6]

Close and secure the vent, restore valve lineup, check for leakage, return controls to automatic and record replacement coolant. Falling reservoir level or repeated makeup after initial gas removal calls for leak and pressure-profile checks. [S1] [S2] [S7]

For U.S. general-industry servicing covered by OSHA’s 29 CFR 1910.147, paragraph (b) distinguishes control commands from energy-isolating devices, while (d)(5) and (d)(6) require stored-energy control and verification before servicing. A zero-speed command therefore does not isolate the loop. Identify electrical supplies, remote restart, pressurized headers and expansion vessels in the equipment-specific energy-control procedure. [S17]

How do you localize a restriction after de-airing?

Follow the pressure-loss path: common filter, header valve, branch hose, QD pair, manifold and complete server path. Bracket each tested boundary with identified taps. Verify compatible, fully seated connectors, actual valve travel and accessible hose geometry before opening a component. [S4] [S5] [S6]

OCP’s Cold Plate Cooling Loop Requirements, Rev 2, ties filtration to sensitive loop components, including connector and microchannel geometry. Use the installed filter/CDU replacement criterion and clean baseline; a borrowed pressure alarm is not a filter specification. Increasing speed can conceal restriction while consuming pressure margin. [S4]

Check quick-disconnect pressure loss and compatibility against the installed pair, including the pressure-test boundary. Preserve debris from an isolated filter instead of forcing it downstream through cold plates.

Gas and contamination can require separate repairs

A filter change can remove resistance and introduce air in the same service event. Record component replacement separately from venting, restore the boundary and remeasure after approved gas removal.

Persistent server-path loss localizes a hydraulic problem without identifying the deposit or authorizing reverse flushing. Route the assembly through its approved service process and use coolant chemistry and contamination evidence to investigate recurrence. OCP’s October 9, 2019 Rev 1.0 cold-plate requirements also support retaining filtration and commissioning history. [S4] [S11]

A rated dry-break disconnection differs from opening a fitting to bleed air. CPC’s undated Everis UQD06/UQDB06 product guidance permits disconnection under pressure for that connector family; the installed coupling limits and server service sequence still govern. Never loosen threaded joints or hold a QD partly engaged as a venting method. OCP’s Rev 2 loop requirements describe QDs as limiting air inclusion, not eliminating it. [S16] [S4]

Worked example: what does restored flow prove?

Restored flow establishes hydraulic delivery; heat duty requires flow and temperature evidence together. Table F applies OCP’s Rev 1.0 reference rating-flow factor to a selected generic 100 kW water loop. OCP’s rating conditions specify water on the FWS and propylene glycol on the TCS; this water calculation uses the flow factor, with its own declared fluid properties. [S5]

Convert volumetric flow to mass flow, then multiply by specific heat and return-minus-supply temperature. Following the sensible-heat method in Caleffi’s idronics 16 (2015), P = ρ × (q/60,000) × cp × ΔT, with P in kW, q in L/min, density in kg/m³ and cp in kJ/(kg·K). [S3]

Method: Anvilfield estimates use Table F’s cited flow factor and declared inputs. Substitute actual coolant properties at mean temperature; concentration and temperature affect thermal duty and hydraulic comparisons.

Why aggregate heat balance is only one part of acceptance

Compare liquid-side heat with the load actually entering that loop: account for residual air cooling, pump heat and energy storage during temperature change. Rack electrical input is not automatically liquid duty, and a bypass can preserve total flow while starving a server branch. [S3] [S4] [S5]

Table F uses a first-order sum for the approximate error allowance and multiplies the lower flow and temperature inputs for the lower duty bounds. Include coolant-property uncertainty and sensor placement or mixing error in the site decision. Under Anvilfield’s proposed lower-bound rule, the nominal thermal pass remains on hold when its lower bound falls below required duty.

Apply the same agreed decision rule to flow: a selected 150 L/min reading bounded by ±2% has a 147 L/min lower bound, so it does not demonstrate a 150 L/min minimum under that rule. OCP’s rating recommendation does not establish any installed server’s minimum. Pair measurement uncertainty with branch-flow and IT-temperature results before applying Table H.

When both flow and temperature rise look adequate, check whether the branch is actually carrying its assigned load. Low flow with a small temperature rise can mean reduced IT demand, a bypass, mixing or a bad sensor; it does not establish spare cooling capacity. A high temperature rise at low flow becomes a repair criterion only when compared with that server’s approved flow and coolant-temperature envelope. [S4] [S5]

What record proves that cooling duty is restored?

Anvilfield’s restored-duty record joins as-found, intervention and as-left evidence; the job’s agreed limits and measurement decision rule govern release. [S2, S4, S5, S6]

Set the stabilization and observation interval before testing, then retain actual start and end times. Branch minimums, pressure limits, coolant and IT temperatures, leak checks and protective controls must pass together at required duty.

The service crew records physical restoration, the controls team verifies overrides and interlocks, and the operating authority releases the load. Include the commissioning witness where the contract requires witnessed acceptance. If service disturbed redundancy, repeat the affected approved failover proof with load protected; use CDU acceptance and redundancy testing for those larger scopes.

Start the acceptance hold after automatic control and the intended valve lineup are restored. An unloaded circulation test closes the gas-removal task; the record remains open for required-duty proof when the specified load is unavailable. Record the tested duty and the outstanding load test explicitly, then keep the affected load within the operating authority’s permitted envelope.

How long does diagnosis take and what belongs in the scope?

Anvilfield estimates 2 to 4 elapsed site hours and 4 to 8 technician-hours for an accessible two-person diagnostic visit with isolation approved and instruments ready. These are November 2026 scope estimates.

Method: allocate 30 to 60 minutes each for records and lineup, measurements, one authorized intervention, and restoration evidence. Multiply elapsed hours by two technicians to estimate staffing.

Define the diagnostic package separately from component replacement. Waiting for access, load authorization, parts, chemistry results or a longer specified duty hold extends the visit. A low-flow diagnostic package specifies branch measurements, one authorized intervention and a restoration record; PorchRocket’s Bid Control tracks public HVAC solicitations where contractors bid that cooling-service scope.

Carry separate allowances for coolant, filters, QDs, waste handling, instruments and load-bank support, using actual quantities and supplier quotes. Include required witness time and controls support explicitly. An inaccessible high point or missing test port changes the scope before the crew starts.

Field checklist and recurrence checks

1. Identify the TCS/FWS boundary and positive- or negative-pressure architecture.
2. Protect the IT load using the approved operating sequence; identify affected branches and the last service/change event.
3. Preserve pre-intervention alarms and trends with a common clock.
4. Confirm the applicable branch flow, temperature and pressure limits.
5. Identify the permitted venting and component-service boundaries.
6. Verify instruments, units, pressure datum and calibration status.
7. Record control mode, pump command, actual speed and overrides.
8. Verify valves, QD engagement and accessible hose routing.
9. Map high points, vents, pressure taps, expansion connection and drain route.
10. Compare filter and branch losses at comparable flow and coolant condition.
11. Confirm offline/load-transfer requirements, authorize model-specific purge or isolated component service, and prepare containment.
12. Record gas release, makeup and any persistent restriction separately.
13. Restore the fluid boundary and verify leak-tightness and valve lineup.
14. Restore automatic controls and alarms; complete affected redundancy proof with load protected.
15. Prove affected-branch flow, required heat duty and IT temperatures over the approved interval using the agreed uncertainty decision rule.
16. Save the as-left evidence, obtain release and schedule event-based recurrence reviews.

Anvilfield recommends recurrence reviews during the first stable post-service run, the next operating shift and the next comparable load period. Compare low-flow events, makeup activity and local pressures at equivalent load and control state. These are event-based checks; scheduled sampling and filter maintenance follow the OEM and coolant specification.

Returning gas directs the investigation toward local pressure, expansion arrangements, ingress, leakage and fill practice. Returning filter loss directs it toward debris origin, coolant chemistry and filter selection. Join the leak-response record to the coolant condition record so repeated bleeding or replacement does not become the default repair. [S4] [S7] [S8]

Review history and how to cite this guide

Byline: Anvilfield. Editorial review date: 2026-11-05. Sources checked: the manufacturer manuals, OCP publications, hydronic references and regulator guidance listed below. Source access: 2026-11-06 UTC.

Update log, 2026-11-05: verified manual editions and purge limits; corrected the QD service boundary and uncertainty method; added a staged gas-plus-restriction diagnosis, measurement checks and acceptance-hold rules.

Cite a specific evidence object using its fragment: fault differential, high-point survey, pressure-flow signatures, purge boundaries or restored-duty record.

How to cite this page: Anvilfield, Liquid-cooling air lock or blockage? A low-flow diagnostic guide, 2026, https://anvilfield.com/field-guides/datacenter/liquid-cooling-air-lock-low-flow-diagnosis/

Related tools

FAQ

People also ask

• What is direct-to-chip liquid cooling?
• What is a CDU in a data center?
• Why do data centers need liquid-cooling leak detection?
• How do I select OCP liquid-cooling quick disconnects for pressure and compatibility?
• What coolant is used in data center liquid cooling?

Sources checked

• [S1] Koolance, ALX-750/1450/2000 User’s Manual English v1.1; printed pp.9 and 15. Filling and priming.
• [S2] Vertiv, CoolChip CDU 121 Installation and Commissioning Guide SL-80277 Rev A, September 2025; §§6.2.4, 6.4 and 6.5. Offline filling, circulation and overrides.
• [S3] Caleffi, idronics 16: Circulation in Hydronic Systems January 2015; Formulae 3-1 and 4-2. Heat transfer, pressure and smooth-pipe loss.
• [S4] OCP, Cold Plate Cooling Loop Requirements Rev 2. Loop components, filtration, fluid properties and instrumentation.
• [S5] OCP, Liquid to Liquid CDU Test Methodology and Performance Rating Rev 1.0; August 2024 cover, initial-release entry 11/01/2024; §3.3.2.4. CDU rating flow and test conditions.
• [S6] OCP, Guidelines to Rack Manifold Requirements and Qualification §3.3. Manifold vents, drains and branch distribution; URL suffix does not establish an edition.
• [S7] Caleffi, idronics 2: Air and Dirt Elimination in Hydronic Systems 2007; p.23. Gas behavior and conventional hydronic top pressure.
• [S8] Caleffi, idronics 32: Troubleshooting Hydronic Systems January 2023. Pressure, expansion and hydraulic troubleshooting.
• [S9] Grundfos, CM self-priming data booklet Document 98514505, issue code 1013; p.10. CM self-priming inlet-pressure calculation.
• [S10] Bürkert, Coolant Distribution Unit (CDUs) Undated application page. Positive-pressure and vacuum architectures.
• [S11] OCP, ACS Liquid Cooling Cold Plate Requirements Rev 1.0, October 9, 2019; §§5.1.2 and 5.1.3. Filter maintenance and commissioning history.
• [S15] Hydraulic Institute, Pump Fundamentals: System Curves Updated July 19, 2024; Pressure and Head Relationship, Major Losses and Minor Losses. Constant-coefficient loss and pressure corrections.
• [S16] CPC, Everis UQD06 and UQDB06 liquid-cooling connectors Undated manufacturer product guidance. Rated dry-break disconnection under pressure; operating conditions remain connector-specific.
• [S17] OSHA, 29 CFR 1910.147: The control of hazardous energy Live regulation; (a), (b), (d)(5) and (d)(6). Scope, isolating devices, stored energy and verification for covered general-industry servicing.